Utilizing Machine Learning to Assess the Impact of Attitudinal, Knowledge, and Perceptual Factors on Diabetes Awareness

doi:10.21203/rs.3.rs-4701414/v1

Download PDF

Research Article

Utilizing Machine Learning to Assess the Impact of Attitudinal, Knowledge, and Perceptual Factors on Diabetes Awareness

https://doi.org/10.21203/rs.3.rs-4701414/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Objectives

The primary objective was to identify and analyze the factors that impact diabetes awareness and perception among diabetic and non-diabetic participants. The study also sought to assess the effectiveness of current health awareness programs and identify gaps in public knowledge about diabetes.

Background

Diabetes poses a significant global health challenge, with increasing prevalence worldwide. Comprehending the behavioral and demographic factors leading to diabetes is important for personalized interventions and prevention strategies in Kuwait.

Methodology:

This study was cross-sectional in nature and employed a quantitative approach. It involved distributing a structured questionnaire to a sample of N = 1268 participants in Kuwait, 391 of them were diabetic and 877 were non-diabetic. The sample was stratified based on age, gender, administrative division and nationality. The study employed machine learning and statistical analyses to examine the nature of the relationship between diabetes awareness and the demographic factors. The study executed a random forest approach before employing a logistic regression model to determine the most significant features influencing diabetes. This involved prioritizing variables based on their importance metrics like a mean dropout loss and mean decrease in accuracy, this ensures that the most important predictors are included in the logistics regression model.

Results

The output shown above describes the results for the logistics regression model indicating the different variables that are significant predictors for diabetes among the participants. From the odds ratio it was observed that age was a significant predictor and people above 60 years of age were 11.47 times more likely to have diabetes compared to the 18–30 age group. For those aged 46–60 the likelihood of having diabetes compared to the 18–30 age group was 5.79 times. Similarly, gender was a significant predictor and males were 2.27 times likely to have diabetes than females. Those who frequently interacted with medical staff were also at higher risk (odds of 1.41), likewise, individuals who had kidney complications were also at higher risk of getting diabetes (odds of 1.60). On the contrast, being overweight decreased the odds of getting diabetic (odds ratio of 0.55), likewise, having pregnancy related diabetes decreased the likelihood of being diabetic (odds ratio of 0.65). From these results, it can be seen that age, gender and certain health complications while interacting with the dependent variable need to be considered while assessing the risk of getting diabetes.

Conclusion

The current study reveals that gender, age groups, kidney disorders and healthcare provider interactions among others, are significantly associated with the awareness and attitude towards diabetes among the Kuwaiti population. On one hand, males and older age groups found to be at higher risk whereas, obesity and pregnancy related diabetes seemed to have a protective effect. The current study findings emphasize the importance of designing specific public health policy and education programs that takes into account the demographic factors to enhance effective diabetes management and prevention strategies. These study findings offer policy knowledge that can assist policymakers to plan and implement more robust health policies that address specific population subgroup needs and challenges.

Diabetes awareness

Public health interventions

Random forest machine learning

Health policy development

Kuwait

Diabetes is a growing global health issue, with a high prevalence, particularly in Kuwait (22%).
Factors like age, gender, obesity, and lifestyle significantly influence diabetes prevalence.
Machine learning algorithms are effective in predicting diabetes and analyzing health data.

ii. What does the paper add to existing knowledge?

This study uses advanced machine learning techniques to analyze diabetes awareness among diabetic and non-diabetic participants in Kuwait.
It identifies key predictors of diabetes awareness, such as age, gender, kidney complications, and interactions with medical staff.
The findings highlight that obesity and pregnancy-related diabetes may lower diabetes risk, contrary to common beliefs.

iii. What insights does the paper provide for informing healthcare-related decision making?

Personal interactions with medical staff are more effective than broad campaigns in raising diabetes awareness.
Age, gender, and health complications are crucial for assessing diabetes risk, guiding targeted interventions.
Integrating machine learning models in public health strategies can enhance the prediction and management of diabetes, informing more effective policies.

Diabetes is a raises global concern because of the increase in its prevalence of the diseases over the years with a global prevalence of about 9.3% in 2019, affecting about 463 million people, this is based on according to International Diabetes Foundation (IDF) data [1]. The prevalence is expected to increase to 10.2% by 2030 and 10.9% before 2045. The number of people living with diabetes expected rise to 700 million during the same period of time (Saeedi et al., 2019). The increase in the prevalence is associated with factors like expansion of population, urbanization, increased aging, poor dietary choices and reduced physical activity [2]. Among the countries that form the Gulf Cooperation Council (GCC) (Saudi Arabia, Kuwait, Qatar, Bahrain, and the UAE), the prevalence of type 2 diabetes (T2D) is very high, the countries are ranked among the top 20 countries globally with the highest rates of diabetes (Arredouani, 2021). In Kuwait, the prevalence of diabetes stands at 22%, this highlights the diseases’ significant burden in the region [3].

Countries in the Middle East like Saudi Arabia have also been greatly impacted by diabetes epidemic, with Saudi Arabia being ranked 5th in the incidence of type 1 diabetes (T1D) and 7th globally in terms of prevalence [4]. In 2017, Saudi Arabia had a 14% prevalence and approximately 7 million reported cases of diabetes [5]. Some of the highest prevalence rates of type 2 diabetes in the world are found in Arabic-speaking countries specifically the middle east [6]. This data highlights the importance of public health measures and specific interventions to address the growing burden of diabetes in these areas.

Kuwait ranks among the highest nations in terms of obesity, with many implications for the prevalence of diabetes as well [7]. Obesity has been proven to be a significant factor in the development of type 2 diabetes, implying that the high prevalence of obesity is likely to also lead to a greater number of diabetic cases in Kuwait. Prediabetes has been proven in many studies to have a connection with obesity in adolescents and children in Kuwait; hence, early detection with prevention strategies is of importance to reduce its prevalence among the population segments in Kuwait [8]. The burden of diabetes has implications other than just health, including economic and societal. Diabetes is related to both mortality and morbidity rates because almost 3 million people die each year from diabetes and complications arising from it globally [9]. In addition to this, diabetes is also an economic burden because of its complications and the cost of managing it that the health system and society at large have to bear. Managing the diabetes epidemic is essential for better health outcomes at an individual level and also for assuring the sustainability of the systems, which further enhances the overall well-being of the population.

The study involves statistical analysis and health data sets to determine the factors influencing the prevalence of diabetes in Kuwait by comparing the diabetic and non-diabetic populations. This research would aim at emphasizing the relationship between the incidence of diabetes and demographic variables to guide effective public health strategies using advanced machine learning tools.

To gain better insight into the determinants affecting diabetes awareness and perception among the participants, factors to consider would be self-care practices, knowledge, levels of education, risk factors, gender differences, illness perceptions, and lifestyle choices. Works by [10, 11] learning approach in the prediction of diabetes. The latter work emphasizes the potential of the machine learning approach in clinical settings as well. Nguyen et al. (2023) Used Random Forest Classifier, Hyper parameters were tuned using grid search. Its performance was evaluated through K-fold cross-validation. This ensured that the model derived high accuracy predictions of diabetes. This study underlined the role of hyper-parameter tuning and cross-validation to increase the predictive power of the models. A study related to [12, 13] reported an accuracy of the model, around 100%. [14] established that the Linear Support Vector Machine and Random Forest models were the best and the latter also emphasized the likelihood of neural network models to have high accuracy. [15] reported a 98%, where the study had used a combination of Support Vector Machine and Random Forest algorithms. [16] also established the potential of machine learning models –LightGBM in accurately detecting diabetes. Another study reported the superiority of the Extra Tree algorithm as a base estimator for the AdaBoost classifier [17]. This study had an accuracy of 90.5%, and this again testifies how useful mutual information-based feature selection approaches and ensemble methods to remove less important features are. This approach goes a long way in improving the performance of tree-based models and testifying the potential application of the models in medical diagnostics.

Kangra & Singh, (2023) used WEKA 3.8.6 to evaluate the performance of machine learning classifiers on datasets obtained from various population segments, including the Pima Indian diabetic (PID) and Germany diabetes datasets [18]. The purpose of this research was to identify the best machine learning model to predict diabetes, given performance matrices and corresponding error rates. They found out that the SVM model was the best when applied on the PID dataset with a 74% accuracy, KNN and Random Forest models when applied on the Germany dataset had an accuracy of 98.7%. [19] and the objective of determining the most efficient machine learning algorithm to be used in developing a diabetes prediction algorithm based on medical data, they compared the performance of several machine learning algorithms, which include, Naive Bayes, Logistic Regression, K-Nearest Neighbor, Decision Tree, Support Vector Classifier and Random Forest. This study, therefore, emphasizes the importance of using significant factors to generate predictive models, the best model can then be used to predict diabetes early using the clinical output. [20] completed a critical review of different machine learning techniques, including Support Vector Machines, Random Forest, Artificial Neural Networks, and conventional algorithms used to make predictions of diabetes mellitus. The differences lie in the datasets used; however, this research has yet again given much attention to the facts of data pre-processing, feature selection, and data cleaning in improving the prediction accuracy. The analysis makes an account of the most effective machine learning models employed on prediction of diabetes, giving emphasis on the deep learning models and thus the higher accuracy when using a large data set. The use of machine learning algorithms for predicting diabetes holds promise in medical research; this will provide robust methodology for the early detection and management of the disease. The uniqueness of the studies is that the machine learning techniques have employed unique strength in handling the prediction of diabetes.

The importance of employing machine learning in predicting diabetes lies in its ability to handle and analyze big datasets efficiently, this reveals the patterns and correlations that may not be revealed using traditional approaches. Several genetic, environmental and lifestyle variables affect diabetes. Machine learning has the ability to manage this complexity effectively this complexity by incorporating data sources like genetic information, health records, and even lifestyle data. Sophisticated methods like logistic regression, support vector machines, and deep learning can be used to identify people at a higher risk, they allow for implementation of prompt intervention practices. The ability to predict future results is important for taking proactive measures use in disease management. This enables healthcare providers to personalize treatment plans and allocate resources efficiently with an aim of improving patient outcomes. Furthermore, machine learning methods consistently get trained and adjust their behaviour in response to the availability of data that has never been seen before (testing data). This improves their ability to make accurate predictions over time. This evidence-based and innovative method for predicting diabetes helps to reduce the impact of the disease and also facilitates the creation of targeted public health plans and policies that aim to mitigate the increasing global effects of diabetes.

The significance of this research is that it effectively examines the variables that impact diabetes awareness and perceptions among diabetic and non-diabetic people in Kuwait. The study employed machine learning approaches to identify the main demographic and behavioural factors that influence knowledge of diabetes, including gender, age, kidney issues, and interactions with medical personnel. While the study highlights trends such as older age groups and the male participants being a more significant risk factor for diabetes, the authors also admit that certain conditions, like obesity and gestational diabetes, may be lowering the risk. These findings bring about the call for targeted public health interventions and individualized educational programs to meet the specific needs of various groups. Using the Random forest technique in conjunction with logistic regression analysis also helps in filtering out relevant factors; the model ensures accurate identification of the relevant components. By providing policymakers with information, this will help the authorities formulate useful health policies aimed at enhancing diabetes treatment and prevention. The primary focus is to reduce the rising burden of diabetes in Kuwait and the surrounding regions.

The aim of the study focuses on using machine learning algorithms to uncover and evaluate the factors that determine diabetes knowledge and perception among the diagnosed and non-diagnosed diabetes population in Kuwait. The primary aim involves using advance machine learning techniques to recognize pertinent demographic, perceptual and attitudinal factors that determine the understanding and perception of diabetes among the different population groups. The study also pursues identifying the most significant factors and evaluating the nature of the association between these attributes and diabetes awareness by Random forest and logistic regression analysis. This helps to attain the purpose of the method that focuses on providing practical and helpful insights that can be used to design targeted public health initiatives and education programs that improve diabetes awareness and its management among different population groups in Kuwait.

This study employed eight disparate machine learning algorithms. These algorithms have been known to be particularly efficient with binomial medical issues. These algorithms include Logistic Regression, Support Vector Machine, Random Forest, Gradient Boosted Trees, Deep Learning, Naïve Bayes, Decision Tree, and the Fast Large Margin variation of Support Vector Machine. The selection of these algorithms was on their precision, resilience, and the ability to manage complex multi-dimensional data that is very common in medical research in a very useful manner. The overall aim of this research is to predict the nature of diabetes or non-diabetes of the subjects based on the chosen variables from the survey data. This project identified these variables as information on demography, lifestyle characteristics, medical background, and attitudes. Each of these variables was imperative as they would determine diabetes awareness and perceptions. This project seeks to develop the most reliable and accurate prediction model using cutting-edge machine learning methodologies. This model is anticipated to provide essential information relating to determinants of diabetes risks and intervention strategies while developing a personalized intervention design. Every machine learning method used in this study had previously been attested to adaptable and, efficient in medical applications.

Support Vector Machine (SVM)

Support Vector Machines (SVMs) have been used widely in medical diagnosis, particularly in categorizing disorders. These models are applied to diagnose breast cancer by scrutinizing mammography images. They are used to determine whether a tumor is malignant or benign by looking at the features of the images. The Fast Large Margin version performs well in handling large datasets with many variables. This ensures accurate classification of the disease as well as quick detection. This, therefore, renders SVM an excellent model for early detection of breast cancer. Support Vector Machines (SVM) has been recognized as an excellent feature in diabetes research and prediction [21]. A good number of studies have been conducted on the effectiveness of SVM models in the prediction of diabetes mellitus and its complications, type 2 diabetes [22–24]. The implementation of SVM in diabetes management and decision-making system has proved to be effective. This indicates the potential of artificial intelligence techniques in clinical practice and self-diagnosis of diabetes [25]. Furthermore, SVM is also utilized to detect diabetes at an early stage along with other models such as Multilayer Perceptron and Stochastic Gradient Boosting [26].

Logistic Regression

Logistic regression has been used in medical research to determine the likelihood of occurrence of a binary outcome based on numerous predictor variables. Typically, logistic regression is used in predicting the chances or the risk of the occurrence of a disease. It is, for example used to determine the likelihood of the occurrence of cardiovascular disease. The likelihood of the occurrence of cardiovascular disease is predicted based on variables, such as the age of the person, blood pressure levels, and levels of cholesterol. This model uniquely and clearly determines which factors significantly predict the chances or the likelihood of a medical condition. Diabetes prediction using logistics regression has generated a lot of recent interest in research. Different researches have been done on the use of supervised machine learning algorithms to predict diabetes. They are primarily aimed at improved accuracy and earlier detection of diabetes. [2] have used decision trees to predict diabetes. Similarly, Qin et al. (2022) used machine learning to predict diabetes, where logistic regression is part of it, using lifestyle variables. The studies point out the relevance of using machine learning for diabetes prediction and logistic regression's level of significance in it. These levels of accuracy have been further improved by smartly improvising this model [27]. [28] has predicted the risk of getting diabetes based on various machine learning algorithms such as logistic regression, random forest, SVM, and XGBoost, with respect to different variables. This study performed a comparative analysis of the algorithms; hence, it explained the strengths and weaknesses of the different algorithms. This proved to be a fantastic way to evaluate the aptness of logistic regression on diabetes prediction. More so, machine learning models have been built using logistic regression, alongside models like CATBoost, XGBoost, and random forests to help in predicting diabetes. These models incorporate the strengths of different algorithms to enhance prediction performance. In the line with the above study, [29] developed logistic regression models with support vector machines, K-nearest neighbors, Naïve Bayes, and random forests in an effort to predict diabetes at an early stage; this indicates that logistic regression can be included in comprehensive prediction frameworks.

Gradient Boosted Trees (GBT)

Gradient Boosted Trees are commonly used in healthcare for predictive analytics such as patient outcomes and the development of a disease. They are applied to predict the chances of readmitting patients with chronic conditions through an analysis of electronic health records. The model can deal with intricate interrelations among variables and is therefore best applicable in understanding multifactorial cases and achieving precise predictions. From clinical research, GBT models, along with other machine learning algorithms like Forest, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), were used to predict diabetes outcomes [30]. Machine learning models were also derived using the Support Vector Machine (SVM), Decision Tree (DT), and Logistic Regression algorithms to predict diabetes, among other diseases. According to [24], ensemble intelligence techniques, such as AdaBoost, GBM, and CatBoost, have thus far been employed in the prediction of diabetes. It uses the low bias and high variance of the decision tree as a base classifier [31].

Deep learning

Deep learning techniques like convolutional neural networks have revolutionized the world of image-based diagnostics. They are primarily employed in analysing medical imaging data such as the identification of diabetic retinopathy in retinal pictures, the classification of skin lesions, and even the detection of any abnormality in the radiographs of a patient. Deep learning models are of a complicated structure and this makes them capable of effectively identifying the minute variations and patterns that are visible in the medical pictures, this leads to the creation of a high level of diagnostic accuracy. Deep learning has proved to be an apt technique for making accurate predictions of diabetes since it possesses the requisite capabilities of handling large biomedically relevant datasets effectively ([32]. It has also been explored through various studies that the use of deep neural networks can also predict diabetes quite accurately ([33]. Also, the effectiveness of different models of neural networks, such as the bidirectional long/short-term memory (BiLSTM) model, has been explored in the accurate prediction of diabetes based on the data of the patient [34]. Also, the studies have focused on employing deep ensemble learning in the provision of a recommendation for an accurate diagnosis in diabetes patient records across different healthcare disciplines [35]. Deep convolutional neural networks and long short-term memory models have been found to be effective in predicting diabetes and preventing associated complications [36]. Also, the merging of deep learning with health care-related data such as retinal images has allowed for the screening and prediction of type 2 diabetes in the patients [37].

Naive Bayes classifiers

Naive Bayes classifiers are popular in the sphere of medical text classification tasks, including the classification of clinical notes, patient diagnosis based on the description of symptoms, and relevant biomedical literature research. Several studies have shown the effectiveness of predicting diabetes using Naive Bayes classifiers. [38] found that in predicting diabetes, the Naive Bayesian classifier attained accuracy of 76.46%, this was better than other classifiers like Decision Tree. Similarly, [39] argued that Naive Bayes classifiers should be employed for the diagnosis of diabetes mellitus, given that this is a probabilistic classification technique. In addition, [40] did a comparison of different machine learning algorithms and observed that Naive Bayes was among the techniques applied to make predictions regarding diabetes mellitus. In addition, [41] argued that this Naive Bayes technique is effective in predicting diabetes at the early stage, which in turn supports health professionals in ensuring early diagnosis. In conclusion, the use of Naive Bayes classifiers in predicting diabetes has been well-documented in many studies, they demonstrate its accuracy and efficiency in diagnosing this chronic condition.

Decision Tree

Decision trees have been frequently used in clinical decision support systems to aid in disease diagnosis and to plan therapy. For example, different clinical standards and patient-specific considerations to select the most suitable treatment for patients suffering from several diseases. The ability to interpret decision trees makes practitioners able to understand the reasoning behind each selection, therefore, it's important to build confidence in the automated systems. Decision Tree algorithms have been widely studies in the prediction of diabetes. Decision Tree models have been used along with other machine learning models such as Naive Bayes, Random Forest, Support Vector Machine (SVM), and Logistic Regression to predict the diabetes [42–46]. All the studies have proven the efficiency of the Decision Tree in picking up features related to diabetes and better prediction with higher accuracy rates, and reported accuracies, for example, around 89.97% in some studies [47]. Decision Trees are particularly useful in combining genetic and clinical characteristics to predict diabetic nephropathy among patients diagnosed with Type 2 Diabetes [21].

Random Forest

Random forests have been used in predicting patient survival rates and disease outcomes. Gene expression dataset are usually analysed in genomics to reveal biomarkers for disorders like cancer. Random forests have an ensemble characteristic that enhances their predictive ability and robustness, this makes them ideal for analysing high-dimensional data often encountered in medical research [48]. The algorithms' capability to handle large datasets including many variables allows for in-depth and precise, modeling of complex medical conditions [48]. Random Forest has been the model of choice in various papers predicting diabetes using machine learning models. [49] investigated the performance of classifiers such as Decision Tree and Random Forest, they found out that the performance of Random Forest is not influenced heavily by dimensionality reduction techniques. Furthermore, Sneha & Gangil, (2019) establish that Random Forest has a high specificity in analyzing the diabetic data, which indicates its potential for accurate predictions. Random Forest has also been used in medical research to predict diabetes early in time. [50] used Random Forest as one of the classification algorithms in their hybrid diabetes prediction model. [51] used Random Forest to predict the possibility of someone getting diabetes, they obtained a high accuracy rate.

Research Setting

This study was conducted in Kuwait in general public areas. The survey questionnaires were randomly distributed to potential respondents who were over 17 years of age. The research personnel visited many public places, including homes, places of work, clinical institutions and shopping malls. In all these areas, the researchers outlined the goals of the study to the administration and volunteers. This research used public sensitization campaigns and targeted educational institutions to ensure that the public was informed and motivated to participate. The study aimed to enhance the level of diabetes awareness among the Kuwaiti population and achieved this by the community involvement and orientation on the goals of the research study. This ensured the achievement of a representative sample and hence reliability and validity.

Data Collection

A well-structured questionnaire was used to collect data used in the research. The questionnaire was developed by a team of diabetes experts who had a good knowledge and competence in the factors under study (appendix 1). This has ensured that the questionnaire has exhaustively addressed all the relevant factors that affect perceptions and awareness of diabetes. The factors include demographic information, lifestyle behaviours, medical background, and understanding of attitudes. The needed sample size was established based on the population of Kuwait, which is about 4.5 million, using power analysis—a very important aspect of the calculation of sample size. The sample size of N = 660 was based on a power of 90% and an effect value of 0.1. However, the research successfully managed to collect data from more than 1200 individuals, which ensures that the analysis generates valid and reliable results. The study only included adults above the age of 17 years, who at the time of the data collection were residing in Kuwait. This ensures that the sample represents the targeted demographic in question. The findings are more reliable and generalizable based on the high sample size. It provides more accurate and detailed insights into the variables that impact the diabetes knowledge and perceptions in Kuwait.

Ethical Consideration

This aspect was crucial in the study to satisfy ethical considerations that ensure privacy and the rights of the participants. The study received ethical approval, which reviewed all protocols and procedures. The study was to the highest degree of ethical consideration and within the principles of autonomy, beneficence, non-maleficence, and justice. The subjects were put to light about the nature of the study using a comprehensive consent statement that outlined the purpose, procedures, potential risks, and benefits of the study. The researchers guaranteed the participants their responses, the data was securely stored, and access was only given to authorized personnel. The participation of the respondents was voluntary, and in case they changed their minds about it, no consequences were to follow. The questionnaire was also carefully designed and no sensitive or intrusive questions were added in regard to the privacy and dignity of the patients. These ethical measures safeguard the integrity of the research; they also secure the well-being of the study's subjects.

Statistical Analysis Procedures

The statistical analysis was done systematically to ensure that the data are well-understood. The first step was to analyze participants' demographic characteristics using frequencies and percentages for categorical variables, and means with standard deviations for continuous variables. The second step was to seek the the univariate associations between the participants' health status (diabetics vs. non-diabetics) and other factors using chisquare tests for categorical variables and t-tests or ANOVA for scale/numerical variables. The third step was dimensionality reduction and the identification of suitable components; then a Principal Component Analysis was performed on the scales of Knowledge (KN), Perception (PER), Attitude (ATT), and Awareness (AW). The fourth step was the use of a machine learning model in the identification of the most important features predicting health status (diabetics vs. non-diabetics). Then the predictors were used in the logistic regression to measure factors that significantly predict diabetes among the participants; they are all on the scales of Knowledge (KN), Perception (PER), Attitude (ATT), and Awareness (AW). JAMOVI software was used to perform descriptive analysis and logistic regression, while RapidMiner was used to generate the output for the machine learning models. This ensured that the significant factors predicting awareness and perceptions of diabetes are presented in a precise and reliable method.

Table 1 present the demographics of the study with sample comprised 1268 participants, having 877 non-diabetics and 391 diabetic patients. χ2 test was conducted to examine the significant differences between diabetic and non-diabetic participants across various demographic parameters. Results of the study showed significant differences across various parameters. Gender distribution showed a higher percentage of females in the non-diabetic group (86.8%) compared to the diabetic group (76.0%). Age group analysis also revealed that a larger proportion of older adults (46–60 years and more than 60 years) were in the diabetic group (37.6% and 48.8%,) compared to the non-diabetic group (30.6% and 19.4%), indicating significant age-related differences between the groups. However, governorate distribution and nationality did not show significant differences, with p-values of 0.567 and 0.243, respectively.

Table 1

Demographic and Clinical Characteristics of Study Participants by Diabetes Status.
	Non-Diabetes (N = 877)	Diabetes (N = 391)	Total (N = 1268)	p value
Gender				< 0.001¹
Female	761.0 (86.8%)	297.0 (76.0%)	1058.0 (83.4%)
Male	116.0 (13.2%)	94.0 (24.0%)	210.0 (16.6%)
Age group				< 0.001¹
18–30 years old	130.0 (14.8%)	9.0 (2.3%)	139.0 (11.0%)
31–45 years old	301.0 (34.3%)	33.0 (8.4%)	334.0 (26.3%)
46–60 years old	268.0 (30.6%)	147.0 (37.6%)	415.0 (32.7%)
More than 60 years	170.0 (19.4%)	191.0 (48.8%)	361.0 (28.5%)
under 18 years old	8.0 (0.9%)	11.0 (2.8%)	19.0 (1.5%)
Governorate				0.567¹
Al Farwaniyah	68.0 (7.8%)	26.0 (6.6%)	94.0 (7.4%)
Al- Ahmadi	65.0 (7.4%)	20.0 (5.1%)	85.0 (6.7%)
Al-Jahra	17.0 (1.9%)	6.0 (1.5%)	23.0 (1.8%)
Capital	464.0 (52.9%)	224.0 (57.3%)	688.0 (54.3%)
Hawally	198.0 (22.6%)	85.0 (21.7%)	283.0 (22.3%)
Mubarak Al-Kabeer	65.0 (7.4%)	30.0 (7.7%)	95.0 (7.5%)
Nationality				0.243¹
Kuwaiti	817.0 (93.2%)	371.0 (94.9%)	1188.0 (93.7%)
Non-Kuwaiti	60.0 (6.8%)	20.0 (5.1%)	80.0 (6.3%)
Note: Data are shown as counts (N) with percentages in parentheses. ¹Pearson's Chi-squared test

The Principal Component Analysis (PCA) with varimax rotation was conducted to identify distinct components, as outlined in Table 2. Component 1, which encompasses variables KN2, KN4, KN5, KN3, and KN1, primarily pertains to knowledge about diabetes, such as familiarity with the concept of diabetes, typical fasting blood sugar levels, pre-diabetes, regular blood glucose tests, and the increasing risk of diabetes in society, with loadings ranging from 0.504 to 0.717 and uniqueness values between 0.474 and 0.630. Component 2 includes ATT1, ATT3, and ATT2, focusing on attitudes toward diabetes information accessibility and the need to enhance diabetes knowledge in the community, with loadings from 0.500 to 0.723 and uniqueness values from 0.424 to 0.644. Component 3, represented by AW1 and AW2, relates to awareness of diabetes prevention and management through diet and physical activity, showing loadings of 0.742 and 0.759, and uniqueness values of 0.409 and 0.437. Component 4 comprises PER2 and PER1, which pertain to perceptions of social media's efficacy in disseminating diabetes information and participation in diabetes awareness activities, with loadings of 0.471 and 0.823, and uniqueness values of 0.302 and 0.723. These components collectively elucidate the multifaceted dimensions of diabetes awareness, knowledge, attitudes, and perceptions within the studied population.

Table 2

Principal Component Analysis Loadings and Uniqueness for Variables Related to Diabetes Awareness and Perceptions
	Component
	1	2	3	4	Uniqueness
KN2	0.717				0.474
KN4	0.663				0.517
KN5	0.582				0.561
KN3	0.535				0.630
KN1	0.504				0.559
ATT1		0.723			0.424
ATT3		0.702			0.447
ATT2		0.500			0.644
AW1			0.759		0.409
AW2			0.742		0.437
PER2				0.823	0.302
PER1				0.471	0.723
Note. 'varimax' rotation was used, KN1: Are you familiar with the concept of diabetes?KN2: Are you aware of the typical range of fasting blood sugar levels?, KN3: Are you familiar with the concept of pre-diabetes?, AWI: Do you believe that diabetes can be prevented?, AW2: Are you aware that maintaining a nutritious diet and engaging in physical activity helps effectively manage diabetes?, KN4: Do you conduct regular blood glucose tests?, KN5: Are you aware of the escalating peril of diabetes transmission in society?ATTI: How do you assess the extent to which information regarding diabetes is accessible in society? ATT2: Are you aware of any medical societies or organizations in Kuwait that offer support and education specifically for those with diabetes? ATT3: Is it necessary to enhance the level of knowledge and understanding regarding diabetes within the Kuwaiti community? PERI: Have you ever engaged in diabetes awareness activities or campaigns? PER2: Is the utilization of social media as a means to disseminate information regarding diabetes considered to be efficacious?

Further, the correlation analysis was conducted (Table 3) to examine the relationships between the constructs of knowledge, perception, attitude, and awareness concerning diabetes, specifically in the context of the effects of social media. Notably, knowledge exhibits a positive correlation with perception (r_p = 0.186, p < 0.001), attitude (r_p = 0.275, p < 0.001), and awareness (r_p = 0.229, p < 0.001). Additionally, perception is positively correlated with both attitude (r_p = 0.117, p < 0.001) and awareness (r_p = 0.091, p < 0.01). Furthermore, a positive correlation is observed between attitude and awareness (r_p = 0.118, p < 0.001). These results indicated the interconnected nature of these constructs and suggest that enhancing knowledge about diabetes through social media can lead to improved perceptions, attitudes, and awareness, ultimately contributing to better management and prevention strategies.

Table 3

Correlation Matrix between Knowledge, Perception, Attitude and Awareness.
	KNOWLEDGE Total		PERCEPTION Total		Attitude Total		Awareness Total
KNOWLEDGE Total	—
PERCEPTION Total	0.186	***	—
Attitude Total	0.275	***	0.117	***	—
Awareness Total	0.229	***	0.091	**	0.118	***	—
Note. * p < .05, p < .01, * p < .001

The comparison of knowledge, perception, attitude, and awareness between diabetic and non-diabetic participants was done using linear model ANOVA (Table 4). The results revealed notable differences. The mean knowledge score for diabetics (4.1 ± 1.2) is significantly higher than that of non-diabetics (3.2 ± 1.4), with a p-value < 0.001, indicating a substantial difference in diabetes-related knowledge between the two groups. However, no significant differences were observed in perception, attitude and awareness between diabetic and non-diabetic participants. These findings suggest that while diabetics possess greater knowledge about diabetes, their perceptions, attitudes, and awareness levels are comparable to those of non-diabetics.

Table 4

Comparative Summary of Knowledge (KN), Perception (PER), Attitude (ATT), and Awareness (AW) Scores Between Non-Diabetic and Diabetic Participants.
	Non-Diabetes (N = 877)	Diabetes (N = 391)	Total (N = 1268)	p value
KNOWLEDGE Total				< 0.001¹
Mean (SD)	3.2 (1.4)	4.1 (1.2)	3.5 (1.4)
Range	0.0–5.0	0.0–5.0	0.0–5.0
PERCEPTION Total				0.304¹
Mean (SD)	0.9 (0.5)	0.9 (0.5)	0.9 (0.5)
Range	0.0–2.0	0.0–2.0	0.0–2.0
Attitude Total				0.198¹
Mean (SD)	2.6 (0.9)	2.6 (1.0)	2.6 (1.0)
Range	1.0–5.0	1.0–5.0	1.0–5.0
Awareness Total				0.430¹
Mean (SD)	1.8 (0.4)	1.8 (0.4)	1.8 (0.4)
Range	0.0–2.0	0.0–2.0	0.0–2.0
¹Linear Model ANOVA

Table 5 shows the results of χ² test that was conducted to examine the significant differences in the health and lifestyle characteristics of the study participants, stratified by diabetes status. The results revealed significant differences in the health and lifestyle characteristics, stratified by diabetes status. The results show that the non-diabetics participants had higher prevalence of gestational diabetes history (42.0% vs 29.9%), family medical history of diabetes (82.2% vs 77.5%), unhealthy eating habits (80.8% vs 76.0%), and overweight/obesity (84.7% vs 76.7%) compared to diabetics’ participants. These findings highlight the importance of targeted interventions focused on improving diet and weight management to mitigate the risk of diabetes.

Table 5

Health and Lifestyle Characteristics of Study Participants Stratified by Diabetes Status
	Non-Diabetes (N = 877)	Diabetes (N = 391)	Total (N = 1268)	p value
Smoking				0.289¹
No	774.0 (88.3%)	353.0 (90.3%)	1127.0 (88.9%)
Yes	103.0 (11.7%)	38.0 (9.7%)	141.0 (11.1%)
Pregnancy diabetes				< 0.001¹
No	509.0 (58.0%)	274.0 (70.1%)	783.0 (61.8%)
Yes	368.0 (42.0%)	117.0 (29.9%)	485.0 (38.2%)
Famil medical history				0.049¹
No	156.0 (17.8%)	88.0 (22.5%)	244.0 (19.2%)
Yes	721.0 (82.2%)	303.0 (77.5%)	1024.0 (80.8%)
Hypertension				0.876¹
No	753.0 (85.9%)	337.0 (86.2%)	1090.0 (86.0%)
Yes	124.0 (14.1%)	54.0 (13.8%)	178.0 (14.0%)
Unhealthy food				0.047¹
No	168.0 (19.2%)	94.0 (24.0%)	262.0 (20.7%)
Yes	709.0 (80.8%)	297.0 (76.0%)	1006.0 (79.3%)
Lack of physical activity				0.799¹
No	298.0 (34.0%)	130.0 (33.2%)	428.0 (33.8%)
Yes	579.0 (66.0%)	261.0 (66.8%)	840.0 (66.2%)
Fat accumulation around the waist area				0.294¹
No	504.0 (57.5%)	237.0 (60.6%)	741.0 (58.4%)
Yes	373.0 (42.5%)	154.0 (39.4%)	527.0 (41.6%)
Overweight/obesity				< 0.001¹
No	134.0 (15.3%)	91.0 (23.3%)	225.0 (17.7%)
Yes	743.0 (84.7%)	300.0 (76.7%)	1043.0 (82.3%)
¹Pearson's Chi-squared test

The prevalence of diabetes-related complications among study participants, as outlined in Table 6, demonstrates significant differences between non-diabetic and diabetic groups. Diabetics exhibit a higher burden of complications compared to non-diabetics. Diabetics have significantly higher rates of kidney complications (74.4% vs. 58.7%, p < 0.001), retinal complications (92.6% vs. 85.2%, p < 0.001), neuropathy (71.1% vs. 60.2%, p < 0.001), and diabetic foot conditions (86.4% vs. 80.5%, p = 0.010). However, heart attack prevalence (31.7% vs. 27.1%, p = 0.096) and brain attack rates (22.0% vs. 21.8%, p = 0.931) are similar between non-diabetic and diabetic groups. These finding highlight the increased complication burden among diabetics, indicating the need for targeted intervention strategies.

Table 6

Prevalence of Diabetes-Related Complications Among Study Participants
	Non-Diabetes (N = 877)	Diabetes (N = 391)	Total (N = 1268)	p value
Kidney complications				< 0.001¹
No	362.0 (41.3%)	100.0 (25.6%)	462.0 (36.4%)
Yes	515.0 (58.7%)	291.0 (74.4%)	806.0 (63.6%)
Retinal complications				< 0.001¹
No	130.0 (14.8%)	29.0 (7.4%)	159.0 (12.5%)
Yes	747.0 (85.2%)	362.0 (92.6%)	1109.0 (87.5%)
heart attack				0.096¹
No	639.0 (72.9%)	267.0 (68.3%)	906.0 (71.5%)
Yes	238.0 (27.1%)	124.0 (31.7%)	362.0 (28.5%)
Brain attack				0.931¹
No	686.0 (78.2%)	305.0 (78.0%)	991.0 (78.2%)
Yes	191.0 (21.8%)	86.0 (22.0%)	277.0 (21.8%)
Neuropathy (such as loss of sensation in the hands or feet)				< 0.001¹
No	349.0 (39.8%)	113.0 (28.9%)	462.0 (36.4%)
Yes	528.0 (60.2%)	278.0 (71.1%)	806.0 (63.6%)
Diabetic Foot				0.010¹
No	171.0 (19.5%)	53.0 (13.6%)	224.0 (17.7%)
Yes	706.0 (80.5%)	338.0 (86.4%)	1044.0 (82.3%)
¹Pearson's Chi-squared test

Table 7 presents the lifestyle choices and health behaviors among participants categorized by diabetes status. Adherence to a healthy, balanced diet is similar between non-diabetics (95.3%) and diabetics (94.1%) (p = 0.364). Regular sports participation also shows no significant difference, with 92.6% of non-diabetics and 90.3% of diabetics engaging in physical activity (p = 0.165). Efforts to reduce weight are also comparable, with 78.1% of non-diabetics and 79.5% of diabetics attempting weight loss (p = 0.566). However, smoking prevalence is significantly higher in non-diabetics (32.5%) compared to diabetics (22.3%) (p < 0.001).

Table 7

Lifestyle Choices and Health Behaviors Among Study Participants Categorized by Diabetes Status
	Non-Diabetes (N = 877)	Diabetes (N = 391)	Total (N = 1268)	p value
Healthy, balanced food				0.364¹
No	41.0 (4.7%)	23.0 (5.9%)	64.0 (5.0%)
Yes	836.0 (95.3%)	368.0 (94.1%)	1204.0 (95.0%)
Do sports regularly				0.165¹
No	65.0 (7.4%)	38.0 (9.7%)	103.0 (8.1%)
Yes	812.0 (92.6%)	353.0 (90.3%)	1165.0 (91.9%)
Reduce weight				0.566¹
No	192.0 (21.9%)	80.0 (20.5%)	272.0 (21.5%)
Yes	685.0 (78.1%)	311.0 (79.5%)	996.0 (78.5%)
No smoking				< 0.001¹
No	592.0 (67.5%)	304.0 (77.7%)	896.0 (70.7%)
Yes	285.0 (32.5%)	87.0 (22.3%)	372.0 (29.3%)
¹Pearson's Chi-squared test

Table 8 highlights the sources of health information among participants with and without diabetes. It shows the distribution of participants who use different sources for health information, including television/radio, daily newspapers/magazines, relatives/friends, and medical staff. The finding shows no significant difference between non-diabetics and diabetics using television/radio (p = 0.157) or newspapers/magazines (p = 0.374). However, non-diabetics are more likely to rely on relatives/friends (42.3% vs. 31.7%, p < 0.001), while diabetics prefer medical staff (61.4% vs. 51.2%, p < 0.001). This indicates that diabetics tend to seek information from medical professionals, whereas non-diabetics more often consult relatives and friends.

Table 8

Sources of Health Information Among Participants With and Without Diabetes.
	Non-Diabetes (N = 877)	Diabetes (N = 391)	Total (N = 1268)	p value
Television/radio				0.157¹
No	557.0 (63.5%)	232.0 (59.3%)	789.0 (62.2%)
Yes	320.0 (36.5%)	159.0 (40.7%)	479.0 (37.8%)
Daily newspapers/magazines				0.374¹
No	744.0 (84.8%)	324.0 (82.9%)	1068.0 (84.2%)
Yes	133.0 (15.2%)	67.0 (17.1%)	200.0 (15.8%)
Relatives/friends				< 0.001¹
No	506.0 (57.7%)	267.0 (68.3%)	773.0 (61.0%)
Yes	371.0 (42.3%)	124.0 (31.7%)	495.0 (39.0%)
Medical staff				< 0.001¹
No	428.0 (48.8%)	151.0 (38.6%)	579.0 (45.7%)
Yes	449.0 (51.2%)	240.0 (61.4%)	689.0 (54.3%)
¹Pearson's Chi-squared test

Predicting Diabetes in Kuwait Using Machine Learning Approaches

Figure 1 demonstrates a machine learning pipeline designed to forecast instances of diabetes by using a range of algorithms. The process starts by importing a dataset from a CSV file that contains pertinent data. The subsequent phase involves acquiring a comprehensive understanding of the dataset by analysing fundamental statistics and facts. The models are trained using chosen columns (features) from the dataset, which are then presented in a tabular style. Subsequently, the data is sent to a data sampler, generating a representative subset for the specific goals of training and testing.

The collected data is inputted into various machine learning algorithms to train predictive models, such as SVM - Fast Large Margin (an optimised Support Vector Machine model), Logistic Regression (a binary classification model), Gradient Boosted Trees (an ensemble learning method that combines multiple weak learners), Deep Learning (neural networks used for learning complex patterns), Naive Bayes (a probabilistic classifier based on Bayes' theorem), Decision Tree (a tree-based model that makes decisions at each node based on feature values), Random Forest (an ensemble method that uses multiple decision trees), and Support Vector Machine (a classification model that finds the optimal hyperplane to separate classes). The trained models were assed using a separate dataset (new dataset) specifically designed for testing the performance of the model. The process generates the evaluation metrics.

The evaluation metrics include the extraction of coefficients of SVM and logistic regression to determine the significance of the features, carrying out statistical analysis on the features, carrying out ROC analysis to assess the classification performance of the models, it also includes generating a confusion matrix that shows the true positives, true negatives, false positives and false negatives. In the last stage, the best model is then used to generate predictions on the testing data in order to predicted the instances of diabetes. The process also includes the use of visualisation tools like Tree Viewer and Pythagorean Tree to analyse the structure of decision trees and random forests, this provides insights into individual trees. This procedure includes data preparation, training the different machine learning models, evaluating their performance, and generating predictions. Each algorithm has unique abilities in predicting the instances of diabetes.

Table 9 shows the performance metrics pf the diverse group of machine learning models used to predict cases of diabetes. The Support Vector Machine (SVM) - Fast Large Margin model had the lowest classification error of 0.259, a standard deviation of 0.035, and a gain of 48. The Fast Large Margin operator employs a fast margin learner based on the linear support vector learning scheme that was proposed by Fan et al., (2008).

The other models considered in the study also offer competitive accuracy with varying trade-offs in training and scoring times. Figure 2 presents the Receiver Operating Characteristic (ROC) curves for various machine learning models employed to predict diabetic cases.

Table 9

Machine learning algorithims adopted to predict the diabetics cases.
Machine Learning Model	Classification Error	Standard Deviation	Gains	Total Time	Training Time (1,000 Rows)	Scoring Time (1,000 Rows)
SVM - Fast Large Margin	0.259	0.035	48.000	6905.000	89.905	250.493
Logistic Regression	0.279	0.016	12.000	5423.000	196.372	220.907
Gradient Boosted Trees	0.282	0.022	10.000	25622.000	407.729	195.266
Deep Learning	0.284	0.016	24.000	8390.000	898.265	228.797
Naive Bayes	0.284	0.024	8.000	9758.000	154.574	790.927
Decision Tree	0.287	0.026	4.000	5476.000	44.164	173.570
Random Forest	0.287	0.022	6.000	25631.000	55.205	299.803
Support Vector Machine	0.290	0.036	10.000	32474.000	337.539	439.842

Table 10 presents the SVM - Fast Large Margin model weights for predicting diabetes cases. The top 10 features influencing diabetes risk are age group (21.23%), history of pregnancy diabetes (16.23%), non-smoking status (16.05%), family medical history (15.42%), overweight/obesity (13.98%), knowledge total (10.79%), gender (9.83%), awareness total (9.63%), retinal complications (6.29%), and hypertension (6.02%). These results highlight the key factors that can inform targeted interventions for diabetes. These top features are used in the logistic regression in the subsequent analysis to identify the diabetes group.

Table 10

Results from SVM - Fast Large Margin – Model weights and important features that predicting diabetics cases.
Attribute	Weight
Age group	21.23%
Pregnancy diabetes	16.23%
No smoking	16.05%
Famil medical history	15.42%
Overweight/obesity	13.98%
KN Total	10.79%
Gender	9.83%
Awareness Total	9.63%
Retinal complications	6.29%
Hypertension	6.02%
Reduce weight	5.56%
Medical staff	5.52%
Governorate	5.45%
Relatives/friends	5.33%
Smoking	4.97%
Neuropathy (such as loss of sensation in the hands or feet)	4.66%
Attitude Total	4.62%
Fat accumulation around the waist area	4.22%
heart attack	4.19%
Diabetic Foot	4.14%
PER Total	2.98%
Lack of physical activity	2.85%
Kidney complications	2.67%
Television/radio	1.34%
Brain attack	1.22%
Unhealthy food	1.21%
Daily newspapers/magazines	0.87%

Table 11 provides the performance metrics for the SVM Fast Large Margin model in predicting diabetes cases. The model achieved an accuracy of 74.1% (± 3.5%), indicating that it correctly classified approximately three-quarters of the cases. The classification error was 25.9% (± 3.5%), reflecting the proportion of incorrect predictions. The Area Under the Curve (AUC) was 76.7% (± 3.6%), demonstrating good overall discrimination ability. Precision was 61.4% (± 12.5%), indicating that about 61.4% of the positive predictions were correct. Recall, or sensitivity, was 50.8% (± 10.8%), showing that the model correctly identified 50.8% of actual diabetes cases. The F-measure was 55.3% (± 10.6%), balancing precision and recall. Specificity was high at 84.9% (± 5.3%), indicating that the model effectively identified non-diabetic cases.

Table 11

Model Performance Metrics for SVM Fast Large Margin
Criterion	Value	Standard Deviation
Accuracy	74.1%	± 3.5%
Classification Error	25.9%	± 3.5%
AUC	76.7%	± 3.6%
Precision	61.4%	± 12.5%
Recall	50.8%	± 10.8%
F Measure	55.3%	± 10.6%
Sensitivity	50.8%	± 10.8%
Specificity	84.9%	± 5.3%

Table 12 presents the results of logistic regression analysis that was conducted to identify significant predictors of diabetic. The variables included in the model were the variables identified as potential predictors based on the results of a machine learning model. The variables included in the model were those that the SVM fast learning model identified as having the highest importance scores, thus ensuring that the most influential factors were considered in the logistic regression. The results of the logistic regression indicate that age is a significant predictor, with older age groups showing higher odds of diabetes compared to the 18–30 years old group: 46–60 years old (OR = 5.4302, p < 0.001), more than 60 years old (OR = 10.8081, p < 0.001), and under 18 years old (OR = 13.0445, p < 0.001). Furthermore, pregnancy diabetes (OR = 0.7180, p = 0.039) and non-smoking status (OR = 0.5432, p < 0.001) are identified as factors significantly reducing the odds of diabetes. Overweight/obesity is also linked to decreased odds (OR = 0.5821, p = 0.009). Conversely, higher knowledge scores are associated with increased odds of diabetes (OR = 1.6673, p < 0.001), and males exhibit higher odds compared to females (OR = 2.3575, p < 0.001). Moreover, enhanced awareness is correlated with lower odds of diabetes (OR = 0.6021, p = 0.003). Seeking advice from medical professionals increases the odds (OR = 1.5109, p = 0.006), while relying on advice from relatives or friends decreases the odds (OR = 0.6973, p = 0.017). Conversely, variables such as family medical history, retinal complications, hypertension, efforts to reduce weight, and governorate did not demonstrate significant associations with diabetes status.

Table 12

Model Coefficients for Binomial Logistic Regression on DM Group.
Predictor	Estimate	SE	Z	p	Odds ratio
Intercept	-3.1881	0.5616	-5.6772	< .001	0.0412
Age group:
31–45 years old – 18–30 years old	0.2286	0.4063	0.5627	0.574	1.2569
46–60 years old – 18–30 years old	1.6920	0.3778	4.4790	< .001	5.4302
More than 60 years – 18–30 years old	2.3803	0.3811	6.2466	< .001	10.8081
under 18 years old – 18–30 years old	2.5684	0.6820	3.7658	< .001	13.0445
Pregnancy diabetes:
Yes – No	-0.3312	0.1603	-2.0667	0.039	0.7180
No smoking:
Yes – No	-0.6102	0.1727	-3.5332	< .001	0.5432
Famil medical history:
Yes – No	-0.1862	0.1899	-0.9804	0.327	0.8301
Overweight/obesity:
Yes – No	-0.5412	0.2083	-2.5978	0.009	0.5821
KN Total	0.5112	0.0646	7.9128	< .001	1.6673
Gender:
Male – Female	0.8576	0.1911	4.4871	< .001	2.3575
Awareness Total	-0.5073	0.1734	-2.9254	0.003	0.6021
Retinal complications:
Yes – No	0.1703	0.2669	0.6379	0.524	1.1856
Hypertension:
Yes – No	0.1193	0.2078	0.5740	0.566	1.1267
Reduce weight:
Yes – No	0.3300	0.2092	1.5776	0.115	1.3910
Medical staff:
Yes – No	0.4127	0.1501	2.7494	0.006	1.5109
Governorate:
Al- Ahmadi – Al Farwaniyah	0.0225	0.4056	0.0554	0.956	1.0227
Al-Jahra – Al Farwaniyah	0.2650	0.6222	0.4259	0.670	1.3034
Capital – Al Farwaniyah	0.0901	0.2895	0.3111	0.756	1.0943
Hawally – Al Farwaniyah	-0.0415	0.3127	-0.1327	0.894	0.9593
Mubarak Al-Kabeer – Al Farwaniyah	0.3605	0.3757	0.9594	0.337	1.4340
Relatives/friends:
Yes – No	-0.3605	0.1512	-2.3842	0.017	0.6973
Note. Estimates represent the log odds of "DM Group = Diabetes" vs. "DM Group = Non-Diabetes"

Overall Background of the Results

The study aimed to assess the impact of attitudinal knowledge and perceptual factors on diabetes awareness among diabetic and non-diabetic participants in Kuwait, utilizing advanced machine learning techniques. A structured questionnaire was distributed to 1268 participants, comprising 391 diabetics and 877 non-diabetics. The primary objective was to identify significant predictors of diabetes awareness and perception, and to evaluate the effectiveness of current health awareness programs. The results revealed that various demographic and behavioral factors, such as age, gender, kidney complications, and interactions with medical staff, significantly influenced diabetes awareness and perception among the participants.

Best Machine Learning Approach for Predicting Diabetes

Among the eight machine learning algorithms tested, the SVM - Fast Large Margin algorithm was considered the best model for predicting diabetes compared to Support Vector Machine (SVM), Logistic Regression, Gradient Boosted Trees, Deep Learning, Naive Bayes, Decision Tree, Random Forest, and SVM - Fast Large Margin. This algorithm had a classification error of 25.9%, with a standard deviation of 3.5%, and a high area under the curve (AUC) of 76.7%. It has the capability of working with large datasets and many features; thus, it is appropriate for this study, which provides accurate disease classification for early detection. In addition, it is robust and very effective with large volumes of data and can accurately determine and analyze the most important predictors.

Study Findings

The analysis using logistic regression was powerful enough to identify predictors of diabetes cases. Age was seen to be a significant predictor because those with more than 60 years were 11.47 times more likely to be diagnosed with diabetes than those with an age of 18–30 years. Those who had ages between 46–60 were 5.79 times more likely to have than those aged 18–30 years. Males were 2.27 times more likely to have diabetes compared to females. Approaching significance was with medical staff (odds of 1.41), whereas kidney complications increased the risk of getting diabetes (odds of 1.60). On the other hand, being overweight decreased the odds of being diabetic(odds ratio of 0.55), similarly, having pregnancy-related diabetes decreased the likelihood of being diabetic (odds ratio of 0.65). These findings highlight the importance of factors like age, gender, and specific health complications in determining the risk of getting diabetes. Moreover, knowledge (KN), perception (PER), attitude (ATT), and awareness (AW) scores were found to be significant factors, with higher knowledge scores being significantly associated with diabetes status. The study also highlighted the role that lifestyle choices like smoking and physical inactivity play in influencing the risk of getting diabetes.

Influence of social media and digital marketing on predicting diabetics

The fact the findings revealed that the predictor "PER1" (engagement in diabetes awareness activities or campaigns) was not a significant factor in the analysis of diabetes awareness is indeed surprising, this analysis is based on the current emphasis on public health education and awareness campaigns. This result suggests that conventional methods of raising awareness, probably including social media campaigns, are not as effective as they are thought to be in influencing health practices or in crating awareness about diabetes within the population in Kuwait.

This is fascinating compared to the significant predictors identified by the analysis, such as the interaction with medical staff and influence from friends and family. These findings highlight the significance of personal interactions and trusted relationships in influencing health awareness and behaviors. The analysis shows that personal, one-on-one interactions is more important in affecting diabetes awareness and management compared to broader public engagement through campaigns.

The fact that PER1" is not statistically significant might raise questions about the content, reach, or engagement strategies of current awareness programs, this includes those conducted via social media. It implies that although social media and other campaign-based approaches have the ability to have a broad reach, their impact on increasing knowledge on health or changing health behaviors might be limited unless they embrace highly targeted or personalized approaches. These insights can be used by policymakers and educators to reevaluate and design more effective strategies for diabetes awareness, probably focusing more on personal interactions and support networks instead of solely relying on broad public health campaigns.

Study Limitations

The study had several limitations despite employing the comprehensive approach. The cross-sectional design limit the ability to infer causality between the identified significant predictors and diabetes awareness and perception. The dependence on self-reported data created biases like recall bias and social desirability bias. Moreover, the study sample, although diverse, may not represent the broader Kuwaiti population, this potentially limits the generalizability of the results. Moreover, the potential confounding variables such as genetic predisposition and in-depth dietary information were not considered in the study. There was also a limitation in the methods of data collection as the study critically relied on face to face measurements that could have influenced the participant's responses due to interview effects.

Recommendations

Based on the findings, several recommendations can be made. The national public health programs of diabetes prevention through an enhanced level of awareness should be designed based on the vulnerability of older adults and males to the condition. These programs should foster improved diabetes awareness, thereby encouraging the public about the need for regular health check-ups and the interactions of patients with medical staff. Special emphasis could be placed on complications such as kidney issues while developing special health campaigns and other forms of health-related educational materials. More so, digital platforms combined with social media can increase the impact and reach of diabetes awareness programs. Policymakers should consider adding machine learning models to public health strategies in order to enhance the effectiveness and accuracy of predicting and managing the course of diabetes. Future studies should consider using longitudinal data in order to get a better insight into the causal relationships between the predictors of awareness and perception of diabetes mellitus.

In summary, demographic and behavioral factors play an important role in determining diabetes awareness and perception among Kuwaiti population. SVM - Fast Large Margin using machine learning algorithms to predict diabetes and identify key predictors contributed to these findings, hence the need for public health-targeted interventions and particular education programs geared towards specific demographic groups. The present study provides insights into the development of more effective diabetes management and prevention strategies by employing advanced machine learning techniques toward mitigating the increasing burden of diabetes in Kuwait. These findings are important for developing future research and public health policies to make sure that the interventions are data-driven and address the needs of the particular population.

Abbreviation	Full Form
IDF	International Diabetes Foundation
GCC	Gulf Cooperation Council
T2D	type 2 diabetes
T1D	type 1 diabetes
PID	Pima Indian diabetic
SVM	Support Vector Machine
KNN	K-Nearest Neighbor
LightGBM	Light Gradient Boosting Machine
GBT	Gradient Boosted Trees
BiLSTM	bidirectional long/short-term memory
DT	Decision Tree

Ethics Approval and Consent to Participate

The study was approved by the Institutional Review Board (IRB) of Kuwait Technical College. Informed consent was obtained from all participants prior to their inclusion in the study. Participants were assured of the confidentiality and anonymity of their responses, with data securely stored and accessible only to authorized research personnel. Participation was entirely voluntary, and participants had the right to withdraw from the study at any time without any consequences.

Consent for Publication

Not Applicable.

Availability of Data and Materials

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request. All data supporting the findings of this study can be made available while maintaining the confidentiality and anonymity of the participants.

Competing Interests

The authors declare that they have no competing interests.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Authors' Contributions

Ahmad T. Al-Sultan contributed to the conception and design of the study. Ahmad Alsaber*, Jiazhu Pan, and Anwaar Al Kandari were responsible for data analysis and interpretation. Balqees Alawadhi and Khalida Al-Kenane contributed to data collection and initial drafting of the manuscript. Sarah Al-Shamali provided critical revisions and important intellectual content. All authors read and approved the final manuscript.

Acknowledgment:

We acknowledge the American University of Kuwait (AUK) for their support and Kuwait Technical College for ethical oversight, both of which were vital to this research. This publication was made possible by the support of the AUK Open Access Publishing Fund.

Conflicts of Interest:

The authors declare no conflict of interest.

Saeedi P, Petersohn I, Salpea P, Malanda B, Karuranga S, Unwin N et al. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: results from the international diabetes federation diabetes atlas, 9th edition. Diabetes Research and Clinical Practice. 2019;157:107843.
Thomas J, Joseph A, Johnson I, Thomas J. Machine learning approach for diabetes prediction. Int J Inform Syst Comput Sci. 2019;8:55–8.
Arredouani A. Greater and more focused measures are needed to tackle diabetes and obesity epidemics in the nations of the gulf cooperation council. Int J Endocrinol. 2021;2021:1–9.
Alaqeel A. Pediatric diabetes in saudi arabia. Int J Pediatr Adolesc Med. 2019;6:125–30.
Salawati E. Mental wellbeing and quality of life among patients with diabetes suffering from hypoglycaemia in saudi arabia. Cureus. 2021. https://doi.org/10.7759/cureus.17586.
Badran M, Laher I. Type ii diabetes mellitus in arabic-speaking countries. Int J Endocrinol. 2012;2012:1–11.
Weiderpass E, Botteri E, Longenecker J, Alkandari A, Al-Wotayan R, Duwairi Q et al. The prevalence of overweight and obesity in an adult kuwaiti population in 2014. Front Endocrinol. 2019;10.
Musher A, Maatook M, Atwan D. Prediabetic risk factors among children and adolescents with obesity. Int J Health Sci. 2022;:10665–74.
Ali N, Akram R, Sheikh N, Sarker A, Sultana M. Sex-specific prevalence, inequality and associated predictors of hypertension, diabetes, and comorbidity among bangladeshi adults: results from a nationwide cross-sectional demographic and health survey. BMJ Open. 2019;9:e029364.
V.Yamuna D, Krishna Chaitanya YD, sri TJ. Diabetes Disease Prediction Using Mach Learn Algorithms. 2022.
Nguyen LP, Tung DD, Nguyen DT, Le HN, Tran TQ, Binh TV, et al. The Utilization of Machine Learning Algorithms for Assisting Physicians in the Diagnosis of Diabetes. Diagnostics. 2023;13:2087.
Kibria HB, Matin A, Jahan N, Islam S. A Comparative Study with Different Machine Learning Algorithms for Diabetes Disease Prediction. In: 2021 18th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE). IEEE; 2021.
Thabit QQ, Fahad TO, Dawood AI. Detecting Diabetes Using Machine Learning Algorithms. In: 2022 Iraqi International Conference on Communication and Information Technologies (IICCIT). IEEE; 2022.
Tsehay A, Assegie P. Nair. The Performance Of Different Machine Learning Models On Diabetes Prediction. 2020.
Alanazi AS, Mezher MA. Using Machine Learning Algorithms For Prediction Of Diabetes Mellitus. In: 2020 International Conference on Computing and Information Technology (ICCIT-1441). IEEE; 2020.
Daniel E, Johnson J, Victor UA, Aditya GV, Sibby SA. An Efficient Diabetes Prediction Model using Machine Learning. In: 2023 4th International Conference on Electronics and Sustainable Communication Systems (ICESC). IEEE; 2023.
Mahedy Hasan SM, Rabbi MF, Champa AI, Zaman MA. A Machine Learning-Based Model for Early Stage Detection of Diabetes. In: 2020 23rd International Conference on Computer and Information Technology (ICCIT). IEEE; 2020.
Kangra K, Singh J. Comparative analysis of predictive machine learning algorithms for diabetes mellitus. Bull Electr Eng Inf. 2023;12:1728–37.
Mahesh TR, Vivek V, Kumar VV, Natarajan R, Sathya S, Kanimozhi S. A Comparative Performance Analysis of Machine Learning Approaches for the Early Prediction of Diabetes Disease. In: 2022 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI). IEEE; 2022.
Shobana G, Bushra SN. A critical review of machine learning approaches for predicting diabetes mellitus. CARDIOMETRY. 2023;:826–33.
Kavakiotis I, Tsave O, Salifoglou A, Maglaveras N, Vlahavas I, Chouvarda I. Machine learning and data mining methods in diabetes research. Comput Struct Biotechnol J. 2017;15:104–16.
Panda M, Mishra D, Patro S, Salkuti S. Prediction of diabetes disease using machine learning algorithms. Iaes Int J Artif Intell (Ij-Ai). 2022;11:284.
Rigla M, García-Sáez G, Pons B, Hernando M. Artificial intelligence methodologies and their application to diabetes. J Diabetes Sci Technol. 2017;12:303–10.
Zou Q, Qu K, Luo Y, Yin D, Ju Y, Tang H. Predicting diabetes mellitus with machine learning techniques. Front Genet. 2018;9.
Contreras I, Vehı́ J. Artificial intelligence for diabetes management and decision support: literature review. J Med Internet Res. 2018;20:e10775.
Kivrak M. Plazma glukoz konsantrasyonu, serum insülin direnci ve diastolik kan basıncı göstergeleri ile makine öğrenme yöntemleri kullanılarak diyabet hastalığının erken tanısı. Med Records. 2022;4:191–5.
Qin Y, Wu J, Wen X, Wang K, Huang A, Liu B, et al. Machine learning models for data-driven prediction of diabetes by lifestyle type. Int J Environ Res Public Health. 2022;19:15027.
Chen S. Comparison of machine learning algorithms and feature visualization analysis for diabetes risk prediction. Journal of Physics Conference Series. 2023;2646:012013.
Poly T, Islam M, Li Y. Early diabetes prediction: a comparative study using machine learning techniques. 2022. https://doi.org/10.3233/shti220752.
Zhang Z, Yang L, Han W, Wu Y, Zhang L, Gao C, et al. Machine learning prediction models for gestational diabetes mellitus: meta-analysis. J Med Internet Res. 2022;24:e26634.
Li D, Liu Z, Armaghani D, Xiao P, Zhou J. Novel ensemble intelligence methodologies for rockburst assessment in complex and variable environments. Sci Rep. 2022;12.
Miotto R, Wang F, Wang S, Jiang X, Dudley J. Deep learning for healthcare: review, opportunities and challenges. Brief Bioinform. 2017;19:1236–46.
Zhou H, Myrzashova R, Zheng R. Diabetes prediction model based on an enhanced deep neural network. Eurasip Journal on Wireless Communications and Networking. 2020;2020.
Rabie O, Alghazzawi D, Asghar J, Saddozai F, Asghar M. A decision support system for diagnosing diabetes using deep neural network. Front Public Health. 2022;10.
Ihnaini B, Khan M, Khan T, Abbas S, Daoud M, Ahmad M. A smart healthcare recommendation system for multidisciplinary diabetes patients with data fusion based on deep ensemble learning. Comput Intell Neurosci. 2021;2021:1–11.
Kumari G, Padmaja P, Suma J. A novel method for prediction of diabetes mellitus using deep convolutional neural network and long short-term memory. Indonesian J Electr Eng Comput Sci. 2022;26:404.
Yun J, Kim J, Jung S, Cha S, Ko S, Ahn Y et al. A deep learning model for screening type 2 diabetes from retinal photographs. 2021. https://doi.org/10.1101/2021.06.29.21259606.
Pradeepika A, Sabitha R. Examination of diabetes mellitus for early forecast using decision tree classifier and an innovative dependent feature vector based naive bayes classifier. ECS Trans. 2022;107:12937–52.
Susilawati D, Riana D. Optimization the naive bayes classifier method to diagnose diabetes mellitus. Iaic Trans Sustainable Digit Innov (Itsdi). 2021;1:78–86.
Faruque M, Asaduzzaman M, Sarker I. Performance analysis of machine learning techniques to predict diabetes mellitus. 2019. https://doi.org/10.1109/ecace.2019.8679365.
Sivakumar S, Venkataraman S, Bwatiramba A. Classification algorithm in predicting the diabetes in early stages. J Comput Sci. 2020;16:1417–22.
Adigun O, Okikiola F, Yekini N, Babatunde R. Classification of diabetes types using machine learning. Int J Adv Comput Sci Appl. 2022;13.
Priyadarsini AJ, Titus R. Survey on predictive analysis of diabetes disease using machine learning algorithms. Int J Comput Sci Mob Comput. 2020;9:19–27.
Sneha N, Gangil T. Analysis of diabetes mellitus for early prediction using optimal features selection. J Big Data. 2019;6.
Xie Z, Nikolayeva O, Luo J, Li D. Building risk prediction models for type 2 diabetes using machine learning techniques. Prev Chronic Dis. 2019;16.
Zou Q, Qu K, Luo Y, Yin D, Ju Y, Tang H. Predicting diabetes mellitus with machine learning techniques. Front Genet. 2018;9.
Bavkar V, Shinde A. Machine learning algorithms for diabetes prediction and neural network method for blood glucose measurement. Indian J Sci Technol. 2021;14:869–80.
Alsaber A, Al-Herz A, Pan J, AL-Sultan AT, Mishra D, Group K. Handling missing data in a rheumatoid arthritis registry using random forest approach. Int J Rheum Dis. 2021;24:1282–93.
Reddy G, Maddikunta P, Lakshmanna K, Kaluri R, Rajput D, Srivastava G, et al. Analysis of dimensionality reduction techniques on big data. Ieee Access. 2020;8:54776–88.
Onyema E, Khalaf O, Tavera C, Tayeb S, Ghouali S, Abdulsahib G et al. A classification algorithm-based hybrid diabetes prediction model. Front Public Health. 2022;10.
Syaputra R, Solichin A. Pregnancy risk level classification using the crisp-dm method. Jurnal Riset Informatika. 2022;5:537–48.

No competing interests reported.

QuestionnaireAppendix.docx

Download PDF

Editor assigned by journal
23 Jul, 2024
Submission checks completed at journal
23 Jul, 2024
First submitted to journal
07 Jul, 2024

You are reading this latest preprint version

Utilizing Machine Learning to Assess the Impact of Attitudinal, Knowledge, and Perceptual Factors on Diabetes Awareness

Status:

Version 1

Abstract

Objectives

Background

Methodology:

Results

Conclusion

Figures

What is already known about the topic?

Introduction

Methodology

Support Vector Machine (SVM)

Logistic Regression

Gradient Boosted Trees (GBT)

Deep learning

Naive Bayes classifiers

Decision Tree

Random Forest

Research Setting

Data Collection

Ethical Consideration

Statistical Analysis Procedures

Results

Predicting Diabetes in Kuwait Using Machine Learning Approaches

Discussion

Overall Background of the Results

Best Machine Learning Approach for Predicting Diabetes

Study Findings

Influence of social media and digital marketing on predicting diabetics

Study Limitations

Recommendations

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1