ML based IoT Framework for Diabetes Detection

doi:10.21203/rs.3.rs-3024165/v1

Download PDF

Research Article

ML based IoT Framework for Diabetes Detection

https://doi.org/10.21203/rs.3.rs-3024165/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

There has been a discernible increase in the prevalence of diabetes in recent years, highlighting the significance of early detection in successfully managing the condition and avoiding complications. A chronic disease called diabetes is characterized by persistently elevated blood sugar levels. This requires consistent monitoring, medication, lifestyle modifications, and adherence to a healthy diet. The system's ability to detect diseases early can help with quick treatment and illness management. The platform makes use of Internet of Things (IoT) technology to provide a simple and quick way to monitor patients' health while reducing difficulties caused by diabetes. Several machine learning algorithms have been used to differentiate between diabetes and non-diabetic patients, including Adaboost, Gridsearch, Evalml, AutoML, and Artificial Neural Network. This research introduces a machine learning model-based IoT system for diabetes detection. The suggested system combines Internet of Things (IoT) devices for gathering physiological data with a cloud-based platform for processing and analyzing the data. It was determined through meticulous investigation that hyperparameter modification greatly improved the performance of the aforementioned algorithms, with the Random Forest algorithm showing the highest accuracy. The IoT-enabled technology offers a trustworthy and affordable option for keeping track of diabetic patients' health, enabling early diagnosis and effective treatment of the condition. Overall, the research's findings highlight the potential for machine learning to improve healthcare outcomes for people with diabetes by shedding light on the role it plays in the detection and management of the disease.

IoT

Diabetes detection

Adaboost

EvalML

AutoML

Artificial Neural Network

Hyper parameter tuning

GridSearchCV

Early detection and monitoring of diabetes is crucial for managing the disease and preventing complications. When the immune system mistakenly attacks and kills the insulin-producing cells in the pancreas, type 1 diabetes, an autoimmune condition, results. Contrarily, type 2 diabetes is a metabolic disorder in which the body either produces insufficient insulin or becomes resistant to it. Effective management of both types of diabetes requires guarding against the issues brought on by chronic hyperglycemia. The most important thing we can do to control either type 1 or type 2 diabetes is to routinely monitor your blood sugar levels[1]. We can use this information to determine the ideal diabetes care plan in cooperation with our medical team. These options allow diabetic complications like heart attack, stroke, blindness, kidney damage, and amputation to be postponed or avoided[1–2]. Early type 2 diabetes diagnosis and management of glycemia and cardiovascular risk factors can have significant positive effects.[2].The U.S. Preventive Services Task Force claims that diabetes early identification and treatment can save lives[3]. Evidence-based medicine advocates for randomized, controlled clinical trials as the gold standard for diabetes care[4].The Internet of Things and machine learning are crucial to the treatment of diabetes. IoT-based approaches can be used to monitor diabetic patients and extract patient data, which can be analyzed using machine learning models.Healthcare professionals may utilize predictive analytics to find trends and forecast outcomes, which could assist them make educated decisions regarding patient treatment.As an illustration, a brand-new system for tracking diabetic patients has been created, which combines predictive analytics with four separate machine-learning algorithms[5].Another research suggested a hybrid strategy that used IoT combined with deep learning networks in order to produce accurate and more precise diabetes forecasts[6].Smart IoT-based gadgets can collect patients' health information in real-time, allowing for the early diagnosis of diabetes[6][5].It has also been suggested to develop a smart medical tracking system that makes use of IoT as well as machine learning to identify diabetes at an early stage[6]. These technologies can help improve diabetes management, reduce complications, and enhance patient outcomes.As IoT and machine learning are increasingly used in healthcare, computers will be able to autonomously produce medical records, predict sickness diagnosis, and continue to keep check of patients in real-time[6]. Machine learning algorithms can be used to analyze huge databases of patient data, which can then be used to identify trends and predict outcomes. This can assist healthcare professionals in making more knowledgeable decisions regarding patient care, which can improve outcomes and save costs[7]. Real-time feedback and alerts can be given to patients via IoT-based remote monitoring devices as they track their blood sugar levels[6]. The research concludes that machine learning models are highly effective in accurately detecting diabetes. The system's early detection capabilities can lead to timely treatment and management of the disease. Further It also provides an affordable and convenient way to monitor patients' health and prevent diabetes-related complications.

Sisodia et al. (2018) aimed to predict diabetes using classification algorithms. The authors used machine learning algorithms, including Naive Bayes, K-Nearest Neighbors, Random Forest, and Support Vector Machine, to analyze medical data of 500 patients. According to their findings, the Random Forest algorithm had the highest accuracy in predicting diabetes, with a score of 76.3%. In 2017, Samant and Agarwal proposed an improvised random forest technique using iris features to diagnose diabetes. They extracted iris features and used soft computing techniques such as fuzzy logic and genetic algorithms to train their model. Their results showed an accuracy of 89.66% in detecting diabetes.Perveen et al. (2020) proposed the use of the Gaussian Hidden Markov Model (GHMM) to handle irregularly sampled longitudinal data and predict diabetes. The authors used medical data from 2,225 patients with diabetes and trained their model using GHMM. Their model achieved an accuracy of 86.9% in predicting diabetes.In a study conducted by Choi et al.,(2019), they used multivariate logistic regression analysis to develop risk scores aimed at predicting the conversion from pre-diabetes to diabetes. During their investigation, they examined the connection between the difference in BMI (Final BMI − Initial BMI) over a 4.7-year observation period and the onset of diabetes. The results show that people with a substantial BMI difference have a higher risk of acquiring diabetes, and losing weight can greatly reduce that risk. A machine learning model was created by Cahn et al. (2020) to forecast the development of diabetes from prediabetes. They used gradient boosting trees and trained their model using demographic, clinical, and laboratory data of 2,531 patients. Their model achieved an accuracy of 0.865 in predicting diabetes progression.In 2021, Deng et al. recommended using deep transfer learning and data augmentation as an approach to enhance the accuracy of glucose level prediction in patients with type 2 diabetes. By utilizing recurrent neural networks and convolutional neural networks, their model was trained on electronic medical information from 132 patients. Their model achieved an accuracy of 89% in predicting glucose levels. In a study conducted by Carter et al. in 2019, they explored the possibility of using predictive modeling and machine learning techniques to diagnose diabetes by analyzing the chemical composition of toenail samples, which can help identify the presence of various components in individuals with diabetes. In their study, Zhang and colleagues (2021) utilized the maximum likelihood ratio approach to eliminate laboratory parameters and collinear variables. For predicting diabetes, they only considered non-invasive characteristics. They used a combination of bagging-boosting and stacking algorithms to complete the classification assignment. Dinh et al. (2019) created a comprehensive strategy using a wide range of indicators to predict cardiovascular disease, pre-diabetes, and diabetes. Data from demographic, nutritional, physical exam, lab, and questionnaire sources were combined as part of a data-driven strategy. The researchers used an ML weighted ensemble model with 131 variables for cardiovascular disease and 123 variables for pre-diabetes and diabetes to categorize the illnesses. In 2020, Severeyn and colleagues examined the area under the glucose and insulin curves of study participants based on their plasma glucose and insulin readings, and researchers evaluated the application of k-means clustering for predicting pre-diabetes and diabetes. In a study by Aminian and colleagues (2020), in order to create risk models for end-organ diabetes complications in two different patient groups, they used machine learning approaches: those who had undergone metabolic surgery and those who had not. Allen and colleagues in 2022, they especially targeted a 5-year risk period, using Gradient Boosted Trees and Random Forest to anticipate the chance of chronic kidney failure as a diabetes consequence. Overall, the findings of these studies suggest that ML models can be effective in detecting diabetes with high accuracy rates. However, there is a need for further validation and testing of these models in diverse populations. storageAdditionally, the integration of multiple data sources such as genetic and environmental factors may improve the accuracy and reliability of these models To sum up, implementing machine learning models for diabetes detection offers a viable path to improving the early diagnosis and management of this illness. IoT is also the core system for accessing and processing the data on line. It collects the data and send it to cloud for further analysis, upendra kumar et.al, 2020.

Table 1

Summary of Research Gaps
Ref.	Authors	Techniques	Year	precision(%)
[8]	Sisodia et al.	Modified machine learning algorithms	2018	76.3%
[9]	Cahn A et al.	Gradient boost	2020	86.5%
[10]	Samant and Agarwal	Improvised random forest	2017	89.66%
[11]	Choi et al.	Logistic Regression, linear and quadratic discriminant analysis, k-Nearest Neighbour	2019	78%
[11]	Perveen et al.	Gaussian Hidden Markov Model	2020	86.9%
[12]	Deng et al.	Deep transfer learning and data augmentation	2021	89%
[13]	Zhang J et al.	Support Vector Machine	2019	90%
[14]	Zhang L et al.	Joint bagging boosting with stacking algorithm	2021	88.5%
[17]	Dinh et al.	Random Forest, gradient boosted tree, logistic regression, SVM and their weighted ensemble mode	2019	94.7%
[18]	Sevenreyn et al	K-means clustering	2020	76%
[19]	Aminiian et al	Multivariable time-to-event	2020	81%
[20]	Allen et al	Regression and random forest	2022	75%

3.1 Proposed model

We proposed a machine learning model approach for predicting diabetes on an Arduino UNO, enabling doctors to access patient data from their smart devices. The system employs smart sensors to gather real-time data from patients, as depicted in Fig. 1. This data is then sent to the Arduino UNO, which uses the information to make predictions about the presence of diabetes in the patient. The predicted output is then transmitted via the cloud to the Healthcare workstation, where it is uploaded to a cloud database server for record-keeping purposes. To ensure that the data remains up-to-date, the result is sent from the Healthcare workstation to the doctor's phone and the real-time monitoring system, respectively. Patients are also able to access their regular details via a connected smart device. This new approach offers a number of advantages over traditional diabetes prediction methods. For one, the use of smart sensors enables the collection of real-time data, which allows for more accurate predictions. Additionally, the system's integration with cloud technology enables doctors to access patient data from anywhere, making it easier to monitor patients and adjust treatment plans as needed. Finally, the system's real-time monitoring capabilities provide patients with regular updates on their condition, allowing them to take a more active role in managing their healthThe main objective of our proposed model is to automatically identify hate speech for large social media data.

3.2 System design of proposed model for accessing data

An IoT based system has been designed to collect data from human beings. This system consist of Arduino UNO for accessing the signals as well as processing the data with sensors and actuators. Further that data will be distinguished using the features of diabetic’s patients. The collected diabetic’s patient’s data is analyzed and sent to physician for proper medication. The physician sent solution to patient on the same electronic devices as well as healthcare workstation. Finally patient’s dataset will be stored in cloud storage which might be accessible by selected users by taking permission from administrator for future analysis .The proposed approach for predicting diabetes using a machine learning model on an Arduino UNO, the system ensures that data is accessible and can be executed through a series of sequential steps. These steps are as follows:

1. Utilization of Smart Sensors: The system incorporates smart sensors to gather real-time data directly from patients. These sensors are capable of collecting various health-related information, including blood glucose levels, heart rate, and other vital signs, as shown in Fig 1. The real-time data collected by these sensors serves as the input for the machine learning model.

2. Processing with Arduino UNO: The collected data from the smart sensors is then transmitted to the Arduino UNO, which serves as the processing unit. The Arduino UNO utilizes the received information to make predictions regarding the presence of diabetes in the patient. Specifically designed machine learning algorithms for diabetes prediction are employed for this purpose.

3. Cloud Transmission: Following the prediction process, the output generated by the Arduino UNO is transmitted via the cloud to a designated Healthcare workstation. This cloud-based transmission ensures secure and reliable transfer of the data over the internet to the intended destination.

4. Healthcare Workstation and Cloud Database Server: At the Healthcare workstation, the predicted output is received and subsequently uploaded to a cloud database server. This server functions as a repository for storing patient data, primarily for record-keeping purposes. The integration of the system with cloud technology allows healthcare professionals convenient access to patient information from any location. This accessibility facilitates effective monitoring of patients and enables adjustments to treatment plans as necessary.

5. Result Distribution: To ensure the data remains up-to-date, the results obtained from the Healthcare workstation are distributed to both the doctor's phone and the real-time monitoring system. This distribution enables healthcare professionals to stay informed about their patients' condition, empowering them to provide timely interventions or make necessary adjustments to the treatment plan.

6. Patient Access: Patients are also granted access to their regular details through a connected smart device. This access encompasses various aspects, including the ability to view their health records, receive updates on their condition, and gain insights into their overall health management. By providing patients with access to their information, they are encouraged to actively participate in managing their own health.

The proposed approach offers several advantages compared to traditional diabetes prediction methods. The utilization of smart sensors allows for the collection of real-time data, enabling more accurate predictions. Furthermore, the integration of cloud technology facilitates seamless access to patient data from any location, thereby enhancing the monitoring and treatment processes. Additionally, the system's real-time monitoring capabilities provide patients with regular updates on their condition, empowering them to play a more active role in managing their health.

Initially raw data has been collected, cleaned and pre-processed. The pre-processed data is spliced for training and testing using different algorithms: Logistic regression, K-Nearest Neighbor, Support Vector Machine, Decision tree, Random Forest, Adaboost, AutoML and EvaML as well as Artificial Neural Network as has been mentioned in Table 2. Further accuracy of the mentioned algorithms has been calculated. Thereafter hyper parameter tuning approach known as GridSearchCV Hyperparameters concepts has been implemented. We are getting improved results in case of all mentioned algorithms, however Tuned Random forest is showing more enhanced result. Same development has been repeated on a different Data set and again similar output has been obtained, which has authenticated this proposed algorithm

4.1 Proposed algorithm - Random forest hyperparameter tuning using GridSearchCV

Input: Datasets & Libraries
Output: Algorithm for Diabetes Prediction
	Begin
	Read Dataset-1 as Pima_Dataset.csv, Dataset-2 as Kaggle_Dataset.csv,d1[ ]as empty, d2[ ] as empty, best algorithm as empty
	Clean the data
	Remove or Fill missing values Split the data for Training and Testing
	For in (Dataset 1, Dataset 2)
	Evaluate the LogisticRegression( ) and store lr_acc at d[i]
	Evaluate the KNN( ) and store knn_acc at d[i + 1]
	Evaluate the SVM( ) and store svm_acc at di[i + 2]
	Evaluate the RandomForest( ) and store rf_acc at d[i + 3] Evaluate the DecisionTree( ) and store dt_acc at d[i + 4]
	Evaluate the AdaBoost( ) and store ad_acc at d[i + 5]
	Evaluate the GridsearchCV and store gscv_acc at d[i + 6]
	Evaluate the AutoML( ) and store aml_acc at d[i + 7]
	Evaluate the ANN( ) and store ann_acc at d[i + 8]
	End For For i in (d1[ ],d2[ ]) max_accuracy _di = max (di[ ])
	End For Count frequency_of_max_accuracy(max_accuracy_d1,max_accuracy_d2) Update the best algorithm End

5.1 Data Collection and Feature Selection

In this instance, we made advantage of a patient data collection. The data set is a matrix of factors or characteristics (features) to be tested in the columns and rows representing the value of the attribute. The datasets of various factors that are provided are PIMA dataset and another from Kaggale.com website greatly assisted us. For machine learning studies on diabetes, one popular dataset is the Pima Indians Diabetes dataset. Data from 768 patients are included in the dataset, along with 8 medical predictor factors and a target variable that indicates whether or not the patient has diabetes. One target variable indicating whether the patient has diabetes and eight medical predictor factors are included in another dataset that contains information on 2000 individuals

One of the main difficulties in training a predictive learning algorithm is the selection of features. In order to limit the number of input parameters for a model, feature selection is a technique that involves choosing only pertinent data and eliminating irrelevant data. It comprises automatically selecting characteristics for your machine learning model that are pertinent to the problem you are attempting to solve. The most important features will be chosen as part of the feature selection process.

Furthermore, by removing or under-representing less important characteristics, reducing training time, and improving learning performance, feature selection contributes to the construction of a more accurate model. Selecting a subset of input features from a dataset that are most pertinent to the issue being solved is the procedure.By decreasing overfitting, enhancing interpretability, and simplifying the model's computational complexity, it is a crucial machine learning phase that can assist to increase the model's accuracy and performance.

5.2 Data Pre-processing

Data preprocessing is the process of modifying or encoding data in order to make it simple for machines to process it and produce reliable results. In other words, content should be altered to make it easier for various algorithms to swiftly comprehend it and generate more precise results. Not all of the pure data must be included in every dataset.Every dataset contains some data in "NULL" type, which makes the dataset redundant and leads to erroneous predictions from the models. In order to overcome these poor accuracy levels and achieve higher and better accuracy levels, data preparation arose. We frequently choose to either eliminate the tuples with incomplete data from the dataset,impute the mean or median values of pertinent columns, or to replace missing values in tuples with missing values, using a different hyperparameter optimization technique.

5.3. Algorithm used for predictions

5.3.1 Logistic Regression

A statistical analysis technique called logistic regression is used to forecast a binary result based on previous dataset observationsWhen the answer variable is categorical, such as true/false, yes/no or pass/fail, this technique is utilised.The chance of an event happening is modeled by logistic regression by making the log-odds for the event a linear mixture of one or more independent factors.For issues with two class values known as binary classification problems, it is a frequently used approach in machine learning..Similar to linear regression, logistic regression models the connection between the independent and dependent variables using a different function.

5.3.2 K-Nearest Neighbor

A non-parametric and supervised learning classifier for classification and regression issues is the k-nearest neighbors (KNN) technique. It is a straightforward method that is simple to use and works by assuming that related objects are nearby. The KNN method makes the assumption that comparable data points are close to one another and uses basic mathematics to represent the concept of similarity.The K-Nearest Neighbor (KNN) algorithm is based on identifying the k training samples that are the closest to a new data point and subsequently predicting the class of the new data point based on the class of the nearest training samples. However, the presence of noisy or irrelevant features, or inconsistent feature scales, may result in reduced accuracy of the KNN algorithm.

5.3.3 Support Vector Machine

To perform classification and regression analysis, a supervised machine learning technique known as Support Vector Machine (SVM) is used. This method works by identifying the best hyperplane that separates the data into different classes. It does this by maximizing the margin between the classes, which is calculated as the distance between the hyperplane and the nearest data points from each class.The SVM method is a potent one that works well when there are more features than samples and can handle high-dimensional data.However, SVM may be computationally costly for big datasets and can be dependent to the regularization parameter and kernel function selection.

5.3.4 Decision Tree

To address classification and regression problems, a method known as decision tree is utilized, which is categorized under the umbrella of supervised learning. This approach involves the creation of a tree-like model of decisions and their potential consequences. Through iterative processes, the model is trained on a labeled dataset to make predictions for new, unseen data. By using decision tree-based supervised learning, accurate predictions can be made for both classification and regression problems. It operates by recursively dividing the data into subsets according to the key characteristic that best distinguishes between the classes of data.The described technique involves building a decision model that takes the form of a tree-like structure. The model represents possible decisions and their respective outcomes, with internal nodes representing attribute tests, branches representing the test results, and leaf nodes representing class labels.Decision trees are capable of handling both categorical and numerical data, making them an ideal choice for data analysis. They are also known for their ease of interpretation and analysis. However, deterministic trees can be highly sensitive to even small fluctuations in the data, which can lead to overfitting of the training data.

5.3.5 Random Forest

The ensemble learning technique known as random forest is commonly used for various tasks, including classification and regression. The random forest algorithm utilizes multiple decision trees to generate predictions, making it a popular and effective technique in machine learning.By building a large number of decision trees and aggregating their predictions, random forest overcomes the propensity of decision trees to overfit to their training set. The robust technique known as random forest,which can handle high-dimensional data, performs well when there are more characteristics than samples. However, with big datasets, random forest can be computationally costly and sensitive to the selection of hyperparameters.

5.3.6 AdaBoost

The famous boosting method known as AdaBoost (Adaptive Boosting) is used in machine learning to merge many weak classifiers into a powerful classifier.The method involves successively training a number of weak classifiers using weighted representations of the training data, with each classifier concentrating on examples that were incorrectly labeled by the preceding classifiers. The weak classifiers are combined with weights based on how accurate they are to create the final classifier.The AdaBoost algorithm is a highly effective method for enhancing the performance of machine learning algorithms, particularly in cases where the individual algorithms being used are considered weak learners. AdaBoost may be computationally costly for big datasets and is susceptible to outliers and noisy data.

5.3.7 GridSearchCV

Finding the ideal set of hyperparameters for a particular model is accomplished via machine learning using the hyperparameter tuning approach known as GridSearchCV.Hyperparameters are variables that cannot be learnt directly from the data and are frequently selected by humans before training based on some intuition or trial and error.In order to determine which collection of hyperparameters produces the highest performance on a validation set, GridSearchCV iteratively searches a specified parameter grid. By identifying the most effective hyperparameters for the issue at hand, it is a potent tool that may assist in enhancing the precision and efficiency of machine learning models.However, GridSearchCV can be computationally expensive for large datasets and high-dimensional parameter grids.

5.3.8 AutoML and EvalML

The creation, training, and assessment of machine learning models are all automated through the process known as automated machine learning (AutoML). AutoML searches for the best model architecture, hyperparameters, and feature engineering techniques to optimize the performance of the model. Automating the machine learning process makes it more user-friendly and accessible. Many of the challenging parts of machine learning, including preventing overfitting and underfitting, unbalanced data, identifying data leaks, and other possible setup-related problems, can be aided by autoML. Model hyperparameter optimization, neural network topology creation, and post-processing of machine learning models are all possible with AutoML.EvalML is a Python library that is available as an open-source tool for automated machine learning. This library provides a high-level API that can be used for the training and evaluation of various machine learning models.. EvalML automates the model selection, hyperparameter tuning, and feature engineering process, and provides a simple interface for evaluating the performance of the models. EvalML is a machine learning library that offers support for a broad range of tasks, including classification, regression, and time-series forecasting..

5.3.9 Artificial Neural Network

The architecture and function of the human brain have inspired the creation of Artificial Neural Networks (ANNs), which are a type of machine learning model. ANNs consist of interconnected nodes organized in layers that analyze data and develop predictive abilities based on the input data..Classification, regression, and time-series forecasting are just a few of the many machine learning tasks that can be performed with ANNs. We can use to automate the process of designing the topology of neural networks, selecting the best activation functions, and optimizing the hyperparameters of the network.

Table 2

Accuracy Table
		Dataset 1	Dataset 2
1.	Machine learning Algorithms
	Logistic Regression	77.3%	77.6%
	SVM	75%	78.8%
	Random Forest	74.24%	92.1%
	K Nearest Neighbour	73.69%	75%
	Decision Tree	73.43%	90.5%
2.	Adaboost	66.40%	65.7%
3.	Grid search cv (Top 3)
	Tuned SVM	77.33%	75.5%
	Tuned KNN	74.47%	91.25%
	Tuned Random forest	78.34%	95.62%
4.	AutoML	77.5%	92.96%
5.	Artificial Neural Network	70.51%	78.5%

5.4. Analysis and Result.

Initially both Datasets are pre-processed as has been shown in Fig. 3. After that concept of confusions matrix on different algorithms are implemented as has been shown in Fig. 4 to Fig. 12. In Table 2, we have presented the outcomes of our research concerning the accuracy of Machine Learning algoriths. Our findings indicate that the Random Forest with hyperparameter tuning using GridsearchCV (referred to as the "Tuned Random Forest"), outperformed in comparison to all other methods with accuracy of 95.62%. On the other hand, the Adaboost method had the lowest performance on both datasets. Additionally, the results of our analysis showed that the best classifier for both diabetes data sets is the Tuned Random Forest. The proposed algorithm is based on Tuned Random forest and has been authenticated through two datasets and in both cases accuracy has been found better in comparison to existing research work.

This research proposes an IoT-based system for diabetes detection that utilizes machine learning models to accurately identify the presence of diabetes. The system's integration with IoT technology allows for efficient data collection and analysis, while the use of machine learning algorithms enables the system to achieve high accuracy in diabetes detection.Specifically, the system employs several algorithms, including Adaboost, AutoML, Artificial Neural Network, and GridSearchCV, to optimize the model's performance. By using these methods, the system can detect the onset of diabetes at an early stage, enabling more effective management and treatment of the disease. The results obtained from the proposed system are promising, indicating that it has great potential for clinical applications. Notably, Tuned Random forest has shown best accuracy on two datasets, which authenticating proposed algorithm. However, it is worth noting that the system's performance may vary depending on the specific features used, and further research may be needed to evaluate its performance on larger and more diverse datasets. .In summary, this research had presented a novel approach to diabetes detection using an IoT enabled system and machine learning models. The system's ability to identify diabetes at an early stage could have significant implications for improving patient outcomes and reducing healthcare costs.

Conflict of Interest: There is no conflict of interest amongst the authors.

Funding Agency: There is no funding agency.

Data Availability: Related data is available.

Monitoring Your Blood Sugar (cdc.gov)
Early Detection and Treatment of Type 2 Diabetes Reduce Cardiovascular Morbidity and Mortality: A Simulation of the Results of the Anglo-Danish-Dutch Study of Intensive Treatment in People With Screen-Detected Diabetes in Primary Care (ADDITION-Europe) | Diabetes Care | American Diabetes Association (diabetesjournals.org)
The Lifesaving Importance of Preventive Health Screenings (usnews.com)
Evidence-Based Diabetes Care | Clinical Diabetes | American Diabetes Association (diabetesjournals.org)
Amine Rghioui, Assia Naja, Jaime Lloret Mauri and Abdelmajid Oumnad An IoT Based diabetic patient Monitoring System Using Machine Learning and Node MCU,Journal of Physics: Conference Series, Volume 1743, The International Conference on Mathematics & Data Science (ICMDS) 2020 29-30 June 2020 Khouribga, Morocco,Citation Amine Rghioui et al 2021 J. Phys.: Conf. Ser. 1743 012035,DOI 10.1088/1742-6596/1743/1/012035
Anum Naseem, Raja Habib, Tabbasum Naz, Muhammad Atif, Muhammad Arif3 and Samia Allaoua Chelloug.Novel Internet of Things based approach toward diabetes prediction using deep learning models Front. Public Health, 24 August 2022 Sec.Digital Public Health Volume 10 - 2022 | https://doi.org/10.3389/fpubh.2022.914106
Navneet Verma, Sukhdip Singh, Devendra Prasad ,Machine learning and IoT-based model for patient monitoring and early prediction of diabetes https://doi.org/10.1002/cpe.7219
D. Sisodia and D. S. Sisodia, “Prediction of diabetes using classification algorithms,”Procedia Computer Science, vol. 132, pp. 1578–1585, 2018
Cahn A, Shoshan A, Sagiv T, Yesharim R, Goshen R, Shalev V, Raz I (2020) Prediction of progression from pre-diabetes to diabetes: development and validation of a machine learning model. Diabetes Metab Res Rev. https://doi.org/10.1002/dmrr.325238
P. Samant and R. Agarwal, “Diagnosis of Diabetes using computer methods: soft computing methods for diabetes detection using iris,” 2017.
Perveen S, Shahbaz M, Saba T, Keshavjee K, Rehman A, Guergachi A (2020) Handling irregularly sampled longitudinal data and prognostic modeling of diabetes using machine learning technique. IEEE 5.https://doi.org/10.1109/ACCESS.2020.2968608,Access 8:21875–2188
Deng Y, Lu L, Aponte L, Angelidi AM, Novak V, Karniadakis GE,Mantzoros CS (2021) Deep transfer learning and data augmentation improve glucose levels prediction in type 2 diabetes patients. NPJ Digit Med. https://doi.org/10.1038/s41746-021-00480-x
Zhang J, Xu J, Hu X, Chen Q, Tu L, Huang J, Cui J (2017) Diagnostic method of diabetes based on support vector machine and tongue images. Biomed Res Int.https://doi.org/10.1155/2017/7961494
Zhang L,Wang Y, Niu M,Wang C,Wang Z (2021) Non Laboratory-based risk assessment model for type 2 diabetes mellitus screening in Chinese rural population: a joint Bagging-Boosting Model. IEEE J Biomed Health Inform 25(10):4005–4016.https://doi.org/10. 1109/JBHI.2021.3077114
Martinez-Millana A, Argente-Pla M, Martinez BV, Salcedo VT, Merino-Torres JF(2019) Driving type 2 diabetes risk scores into clinical practice: performance analysis in hospital settings. J Clin Med. https://doi.org/10.3390/jcm8010107
Mohan V, Deepa R, Deepa M, Somannavar S, Datta M (2005) A simplified Indian Diabetes Risk Score for screening for undiagnosed diabetic subjects. J Assoc PhysIndia 53:759–763
Dinh A, Miertschin S, Young A, Mohanty SD (2019) A data-driven approach to predicting diabetes and cardiovascular disease with machine learning. BMC Med Inform Decis Mak. https://doi.org/10.1186/s12911-019-0918-5
Severeyn E, Wong S, Velásquez J, Perpiñán G, Herrera H, Altuve M,Díaz J (2020) Diagnosis of type 2 diabetes and pre-diabetes using machine learning. In: IFMBE proceedings, vol 75. Springer, https://doi.org/10.1007/978-3-030-30648-9_105,pp 792–802.
Aminian A, Zajichek A, Arterburn DE et al (2020) Predicting 10-year risk of end-organ complications of type 2 diabetes with and without metabolic surgery: a machine learning approach. Diabetes Care. https://doi.org/10.2337/dc19-2057s,pp-43(4):852–859
Allen A, Iqbal Z, Green-Saxena A et al (2022) Prediction of diabetic kidney disease with machine learning algorithms, upon the initial diagnosis of type 2 diabetes mellitus. BMJ Open Diabetes Res Care 10:e002560. https://doi.org/10.1136/bmjdrc-2021-002560
Choi BG, Rha SW, Kim SW, Kang JH, Park JY, Noh YK (2019)Machine learning for the prediction of new-onset diabetes mellitus during 5-year follow-up in non-diabetic patients with cardiovascular risks. Yonsei Med J 60(2). https://doi.org/10.3349/ymj.2019.60.2.191,pp-191
Upendra Kumar,Shreyasi Uphdhyay, Neeraj Kumar,Subhash Chandra Pandey(2020) Isolation of DDoS Attack in IoT : A New Perspective, Wireless Personal Communications, pp. 2493-2510

Download PDF

Reviewers agreed at journal
01 Aug, 2023
Reviewers invited by journal
30 Jul, 2023
Editor assigned by journal
09 Jun, 2023
First submitted to journal
05 Jun, 2023

You are reading this latest preprint version

ML based IoT Framework for Diabetes Detection

Status:

Version 1

Abstract

Figures

1. Introduction

2. Related works

3. Proposed model and System Design

3.1 Proposed model

3.2 System design of proposed model for accessing data

4. Proposed Methodology for Diabetic detection:

4.1 Proposed algorithm - Random forest hyperparameter tuning using GridSearchCV

5. Simulation Results

5.1 Data Collection and Feature Selection

5.2 Data Pre-processing

5.3. Algorithm used for predictions

5.3.1 Logistic Regression

5.3.2 K-Nearest Neighbor

5.3.3 Support Vector Machine

5.3.4 Decision Tree

5.3.5 Random Forest

5.3.6 AdaBoost

5.3.7 GridSearchCV

5.3.8 AutoML and EvalML

5.3.9 Artificial Neural Network

5.4. Analysis and Result.

6. Conclusion

Declarations

References

Status:

Version 1