Applications of Hierarchical Classification Techniques for Classifying Anxiety Among Undergraduate Admission Candidates in Bangladesh

doi:10.21203/rs.3.rs-4713624/v1

Download PDF

Article

Applications of Hierarchical Classification Techniques for Classifying Anxiety Among Undergraduate Admission Candidates in Bangladesh

https://doi.org/10.21203/rs.3.rs-4713624/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

The undergraduate entrance exam, which is required for admission to either Bangladesh's public higher education institutions or medical institutions, is one of among the most important investigations in a student's life. The purpose of the current research was to employ sophisticated machine learning techniques to determine clinical anxiety prevalence among Bangladeshi admission participants while additionally discovering associated risks.

Methods

A total of 5239 individuals were randomly sampled and surveyed using the General Anxiety Disorders Scale (GAD-7) to assess the prevalence of anxiety. Boruta found anxiety prevalence predicting factors. We evaluated the decision tree (DT), support vector machines (SVM), random forest algorithm (RF), and extreme gradient boost (XGBoost) using traditional classification (TC) as well as hierarchical classification (HC), and their performance was evaluated using parameters of Confusion matrix, ROC curves, and the cross-validation.

Results

Among the respondents, one-third of them reported a severe level of anxiety. Participants' family problems, drug addiction, and eleven more were selected as risk factors predicting anxiety by using Boruta. The performance was tested based on two different classification techniques, considered traditional classification and hierarchical classification. Overall, the hierarchical classification in terms of local classification of the xtreme gradient boosting model (Accuracy = 0.926, Sensitivity = 0.987, Specificity = 0.22, F-score = 0.963, and AUC = 0.71) performed better and authentically predicted anxiety.

Conclusion

The findings offer legislators, stakeholders, and household members an opportunity to address this significant crisis proactively through enhanced policy, concentrated psychological well-being promotion, and the development of extremely effective mental health services.

Biological sciences/Psychology

Health sciences/Health care

Anxiety

Feature selection

Hierarchical classification

Cross-validation

A rising body of research indicates an increase in the number of cases of psychological illnesses among students in particular those attending secondary secondary school including institutions of higher learning (Campbell et al., 2022). The journey from secondary school to university involves a slew of changes across many different areas, especially social, educational in nature, and emotional components of advancement (Faye-Dumanget et al., 2017). Individuals who have completed their secondary school education in many countries, including Bangladesh, India, Russia, France, and Germany, are required to take an exam for admission to higher schooling, and obtaining a higher education is recognized as an outstanding accomplishment, as it allows those who possess creative structures and competencies to gain access to a tremendous the bodies of comprehension (Mamun, Misti, et al., 2022).

The undergraduate admission test is an important turning point for students in Bangladesh, demonstrating the first step of their professional aspirations. With only a few seats available, meeting the eligibility criterion isn't enough to gain admittance. This means that students must face the admission examinations head-on and perform magnificently to be able to secure admission to an educational institution (Mamun, Misti, et al., 2022). For example, in the year of study 2021-22, almost one million students throughout the country took the Upper Secondary School Certificate and identical tests, with 189,169 gaining an overall label of 5 (Over 95 % of Students Clear HSC, Equivalent Exams, 2022). These restricted vacancies put sigificant anxiety on enrollment candidates (Protikuzzaman et al., 2020). Most applicants fall short it, which means that as a result, they lose entrance. Plenty of them contemplate their own because of being prohibited entrance to their desired university (Protikuzzaman et al., 2020). According to a statewide well-being college research project executed by the National Mental Health Association, approximately one out of every ten learners in higher education has been evaluated for anxiety (Anderia, 2002). Compared to the American College and Graduate Wellness Association's 2014 survey, approximately fifty percent of the collegiate demographic complained of greater than normal levels of anxiety during the previous year, with certain individuals describing it as dreadful (Wald et al., 2014). Furthermore, prolonging psychological care would assist nations around the globe to achieve SDG 3 (well-being) by 2030, which seeks to avoid or alleviate diseases that are not transmissible, minimize deaths from preventable causes by one-third, and encourage psychological wellness and health (UN, n.d.). A lot of studies have investigated the occurrence of mental wellness difficulties among students at colleges and universities, and their results show that an extensive percentage of students attending colleges worldwide are experiencing mental disorders (Adewuya, 2006; Bayram & Bilgel, 2008; Hassan et al., 2020; Kulsoom & Afsar, 2015; Mamun, Hossain, et al., 2022; Nordin et al., 2009; Ovuga et al., 2006; Saleem & Mahmood, 2013; Seim & Spates, 2009; ul Haq et al., 2018; Verger et al., 2010).

Still, very few investigations have researched the mental well-being struggles of entrance aspirants. According to a major study on the subject, depression, anxiety, and stress constitute the most frequently received emotional wellness concerns among Bangladeshi students in higher education, with incidences exceedingly high as 54.3%, 64.8%, and 59.0% (Abu Hena Mostafa Alim et al., 2014; Mamun et al., 2021; Mamun, Hossain, et al., 2022; Sweatt,S.K, Gower, B.A, Chieh, A.Y, Liu, Y, Li, 2016). Another study conducted among students in colleges, according to a thereafter Covid-19 investigation from Bangladesh that 75 percent those who were polled fulfilled the minimum requirement for depression, while 60 percent exceeded the threshold indicated anxiety (Siddik et al., 2024). According to the survey, 74 percent of university admission seekers suffered from depression, while 26 percent possessed moderate depressive disorders, a quarter encountered fairly severe depressive disorders, and 22 percent faced severe depressive disorders (Siddik et al., 2023).

All previous investigations on anxiety disorders in Bangladesh have been overly focused on either school-college or undergraduate medical-university students. Most anxiety-related studies used the method of logistic regression (LR) model to predict the occurrence of psychiatric disorders. Although Hierarchical Classification (HC) is a popular machine learning (ML) model for classification when hierarchy is present in datasets, we want to compare the performance of various ML models that use hierarchical classification, such as SVM, DT, RF, and XGBoost as to predict the occurrence of anxiety among Bangladeshi university entrance applicants. In general terms, machine learning in medical settings attempts to simulate some medical outcomes implementing an assortment of inputs (Mateen et al., 2020; Roberts et al., 2021).

The potential of artificial intelligence (AI) for healthcare applications is significant as demonstrated by the recent release of instances that demonstrate ML-based technologies exhibiting human-level or higher detection and predicting abilities throughout literally every single specialty (Topol, 2019). The ML strategy could investigate into other appropriate information on this significant health-related matter. As a result, we are inspired to determine hazards (features) and anticipate the incidence of anxiety among Bangladeshi University entrance candidates through HC. The goals of the study were to determine the incidence of exposure to potentially anxious circumstances among respondents, to discover the held accountable characteristics using an advanced machine learning procedure, to apply conventional machine learning (ML) and hierarchical techniques, to compare the techniques utilized for predicting depression and anxiety in measuring the effect of covariates, and to identify and predict the best machine learning approaches.

Ethical clearance

The Institutional Review Board (IRB) of Noakhali Science and Technology University granted the study ethical clearance and logistical support. The provided reference code is NSTU/SCI/EC/2022/89(A). This research was carried out in accordance with the principles outlined in the Declaration of Helsinki. The study included participants who provided written informed consent to participate, whereas those who did not provide consent were excluded from participation to fill up the survey form.

Study design and data collection procedure

A cross-sectional research study following random simple sample procedures was done throughout each of the administrative units of Bangladesh between the month of April 2022 to the end of July 2022, shortly after the last surge of COVID-19 in Bangladesh. The data has been gathered from pupils who had taken the University entrance exams for the course of study in the period 2021-20. Due to budgetary limitations, data has been gathered Internet via Google Sheets, a straightforward and fast electronic questionnaire. The questionnaire contained socio-demographic as well as health-related questions for patients (GAD-7) components. The GAD-7 measures are considered legitimate and trustworthy for this collection of data, with a Cronbach's alpha rating of 0.78. To determine both the feasibility and usefulness of the studies, an initial experiment with a total of 51 pupils was conducted. They were subsequently overlooked throughout the resulting examination of the data. While the final investigation study took place, the survey questionnaire underwent some rounds of modifying to guarantee that it matched all the specifications.

Sample size and sampling technique

The Sample size was calculated as follows: sample size is being (((z^2)*p*(1-p))/(e^2))/(1+((z^2)*p*(1-p))/(e^2*N)) = 4211 In other words, N is the number = 1.4,000,000 is the aggregate amount of individuals who scored the university entrance exams; Z = 1.96; Z benefit to 95 percent confidence intervals; P = Considered occurrence (i.e., fifty percent); e = Preferred precision (e.g., 0.02). If 19.61% of the participants are unresponsive (80.39% respond), the final sample number is n = 4211/0.8039 = 5239 employing the straightforward/convenience sampling procedure.

Criteria for inclusion and exclusion

The following were the eligibility criteria employed in this study: i) Students requesting entrance in a undergraduate program who passed the higher secondary school certificate (HSC) exam throughout the 2021-22 year of study. ii) Respondents possessing Bangladeshi citizenship who are submitting applications for entrance to a university. The following were the removal criteria employed in this inquiry: i) Individuals registered in courses of study where English has been employed as the major medium of education; ii) Individuals enrolled in Career paths in vocational iii) Individuals who took education in another country In the Bangladeshi education.

Measures

Socio-demographic determinants

Socio-demographic variables were encompassed age category, sex, marital status, household financial situation (a lower class, the lower middle class, the middle center class, middle-upper group, upper group), use of social media platforms, religion, any kinds of dependence on drugs, pre-marital relationship, sexual behavior, conventional exercise, time spent in the study (below three hours, three to four hours, four to five hours, five to six hours, above six hours), execute religion, smoking status, Blackmail.

General Anxiety Disorder

The General Anxiety Disorder was utilized to assess the level of anxiousness encountered by the individual who responded General Anxiety Disorder (GAD-7). The GAD-7 is one amongst the greatest commonly used and user-friendly assessments for identifying anxiety. For GAD-7 questions, the codes range from "0 = not at all" to "3 = almost every day". The severity of anxiety was classified into four groups based on scores ranging from 0 to 4, 5 to 9, 10 to 14, and more than 15, as follows: none-minimal, mild, moderate, and severe (National HIV Curriculum, n.d.). In this study, moderate to severe was deemed the prevalence of anxiety (anxious), and none-minimal to mild as not anxious was developed for the first layer (Hosen et al., 2021). None-minimal, mild, moderate, and severe was developed for the second layer.

Statistical Analyses

The purpose of this investigation had been to identify factors associated with risk alongside predicting anxiety among Bangladeshi undergraduate admission candidates utilizing decision trees (DT), random forests (RF), support vector machines (SVM), and xtreme gradient boosting (XGBoost). Following data collection, all the data was processed cleaned on a regular and addressed for the final assessment using Microsoft Excel 19. The Microsoft Excel spreadsheet was then imported into R programming software, where inferential and descriptive data analysis (Chi-square) were initially employed. Our approach encompasses appropriate collecting of data and initial processing, picking out features (the risk factors) by employing the Boruta approach, a dividing the full collection of data into both test and training information sets employing machine learning techniques in a training set and assessing their effectiveness on the test data sets—and ultimately employing the machine learning model that accomplished the best to predict anxiety based on a complete information set. Four metrics of performance from the confusion matrix have been employed for assessing the performances: sensitivity, specificity, and accuracy, as well as the ROC curve and cross-validation. Every aspect ML technique with flat and local classification were created in Python using the scikit-learn for the entire package. To pick the risk factors, the Boruta algorithm has been implemented in the programming language R through the "Boruta" software package.

Selection of features

Figure 1 shows that, using the Boruta approach, twelve features (practice religion, pre-marital affairs, study time, SSC result, gender, drug addiction, blackmail, Sickness, exercise, exam preparation, and family problem) were picked from among twenty-one study variables as the responsible features to predict anxiety among undergraduate admission candidates. These twelve factors were subsequently employed for assessing the effectiveness of ML techniques and to classify their anxiety.

(Fig. 1)

Descriptive analysis and cross-tabulation of the selected feature

Figure 2 depicts the donut chart which explores the prevalence of anxiety, 60% of respondents were anxious and 40% of respondents were not anxious.

(Fig. 2)

The frequency analysis of the study's responsible features gives an overview of the respondents of the study data where we found that entire of 5239 pupils were eligible for the study who had data on whether their anxiety was minimal, mild, moderate, or severe cases. 16.89% of them had their anxiety at a mild level, 23.40% of them had their anxiety at a moderate level, 22.96% of them had their anxiety at a moderately severe level and 36.74% of them had their anxiety at a severe level. Out of 5239 respondents, 62.35% were male group remaining 33.64% were female, lower class (17.46%) income level followed by upper class (20.80%), 25.35% were drug addicted remaining 74.65% were not addicted, 12.97% were doing exercise remaining 87.02% were not, 22.06% were committed pre-marital affairs remaining 81.75% were not committed, 20.36 were below three hours followed by 37.25 were above six hours, 88.96% were practiced religion remaining 11.04% were not, 8.99% were blackmailed remaining 91% were not blackmailed, 44.51% respondents who have a family problem remaining 55.48% were respondents who haven’t a family problem, 13.22% respondents who have sickness remaining 86.77% were not sicked. 0.133% were under 3 SSC results followed by GPA 5 was 57.33%. 9.6% of respondents had mental problems remaining 90.32% were not. 10.74% of respondents were not confident about exam preparation followed by 9.4% who were more confident about exam preparation as shown in Table 1.

Table 1

Frequency distribution and relationship with anxiety level among the undergraduate admission candidates.
Features	Level	Anxiety Level (n = 5239; Severe = 36.74%)					Chi square test
		Total					Chi square test
		n	Minimal	Mild	Moderate	Severe	\(\:{{\chi\:}}^{2}\)	p-value
		(%)	(%)	(%)	(%)	(%)
Gender	Male	3267	457	696	757	1357	114.08	< 0.001***
	Male	(62.3)	(14.0)	(21.3)	(23.2)	(41.5)
	Female	1972	428	530	446	568
	Female	(37.7)	(21.7)	(26.9)	(22.6)	(28.8)
Drug Addiction	Yes	1328	112	264	310	642	150.3	< 0.001***
		(25.3)	(8.4)	(19.9)	(23.3)	(39.2)
	No	3911	773	962	893	1283
		(74.6)	(19.8)	(24.3)	(22.6)	(34.5)
Exercise	Yes	680	191	209	141	139	134.152	< 0.001***
		(12.9)	(28.1)	(30.7)	(20.7)	(20.4)
	No	4559	694	1017	1062	1786
		(87.1)	(15.2)	(22.3)	(23.3)	(39.2)
Premarital Affairs	Yes	1156	122	234	282	518	69.19	< 0.001***
		(22.1)	(10.6)	(20.2)	(24.4)	(44.8)
	No	4083	763	992	921	1407
		(77.9)	(18.7)	(24.3)	(22.6)	(34.5)
Study Time	< 3 hour	1067	119	205	215	528	145.33	< 0.001***
		(20.3)	(11.2)	(19.2)	(20.1)	(49.5)
	3–4 hour	663	92	164	150	257
		(12.6)	(13.9)	(24.7)	(22.6)	(38.8)
	4–5 hour	709	114	161	199	235
		(13.5)	(16.1)	(22.7)	(28.1)	(33.1)
	5–6 hour	848	138	199	213	298
		(16.2)	(16.3)	(23.5)	(25.1)	(35.1)
	> 6 hour	1952	422	497	426	607
		(37.3)	(21.6)	(25.5)	(21.8)	(31.1)
Practice Religion	Yes	4661	830	1132	1069	1630	71.21	< 0.001***
		(88.9)	(17.8)	(24.2)	(22.9)	(35.0)
	No	578	55	94	134	295
		(11.1)	(9.5)	(16.3)	(23.3)	(51.0)
Blackmail	Yes	471	24	66	100	281	136.775	< 0.001***
		(9.0)	(5.1)	(14.0)	(21.2)	(59.7)
	No	4768	861	1160	1103	1644
		(91.0)	(18.1)	(24.3)	(23.1)	(34.5)
Family Problem	Yes	2332	129	405	536	1262	731.73	< 0.001***
		(44.5)	(5.5)	(17.4)	(23.0)	(54.1)
	No	2907	756	821	667	663
		(55.5)	(26.0)	(28.2)	(22.9)	(22.8)
Sickness	Yes	693	57	116	146	374	116.40	< 0.001***
		(13.2)	(8.2)	(16.7)	(21.1)	(54.0)
	No	4546	828	1110	1057	1551
		(86.8)	(18.2)	(24.4)	(23.3)	(34.1)
SSC Result	GPA 3	7	2	0	1	4	FET= 146.37	< 0.001***
		(0.13)	(28.6)	(0.0)	(14.3)	(57.1)
	3 to less 4	248	62	61	64	57
		(4.6)	(24.9)	(25.3)	(26.1)	(23.7)
	4 to less than 5	1987	427	538	427	597
		(37.9)	(21.5)	(27.10)	(21.5)	(29.9)
	GPA-5	3004	594	895	940	1602
		(57.3)	(14.7)	(22.2)	(23.3)	(39.7)
Mental Problem	Yes	504	34	89	110	274	88.79	< 0.001***
		(9.7)	(6.7)	(17.6)	(21.7)	(54.0)
	No	4732	851	1137	1093	1651
		(90.3)	(18.0)	(24.0)	(23.1)	(34.9)
FET Fisher’s Exact Test.

(Table 1)

Flat Classification Technique

Flat classification revealed that anxiety was divided into four different categories as a root node minimal, mild, moderate, and severe level. This flat classification technique was used to classify the root node of anxiety. Figure 3 depicts the proposed formulation of a flat classification technique for anxiety.

(Fig. 3)

The different non-ensemble (DT and SVM) and ensemble methods (RF and Xtreme Gradient Boosting) were used to classify the root node of anxiety. Employing 70 percent of the collected samples as trained data as well as 30 percent as test data, we determined the mean level of the degree of sensitivity, specificity, then accuracy, and precision of the DT, SVM, and RF, and XGBoost computational methods for estimating anxiety among undergraduate admission seekers, and the outputs are provided as follows.

(Table 2)

Table 2

Accumulated performance for the flat classification
Indicators	DT	SVM	RF	XGBoost
Accuracy	0.411	0.422	0.425	0.300
Sensitivity (average)	0.352	0.351	0.388	0.477
Specificity (average)	0.792	0.793	0.801	0.776
Prevalence (average)	0.249	0.249	0.249	0.249
Positive Predictive Value (average)	0.321	0.370	0.384	0.327

The accumulated performance of the flat classification techniques is displayed in Table 2. According to the result of many algorithms used to classify anxiety using test data. This table shows the very wired performances that are unaccepted to classify anxiety. So, we have to apply the local classification technique.

(Fig. 4)

Figure 4 represents the ROC curve for Anxiety as a response variable. This ROC curve represents how much the False Positive Rate which is called Specificity and True Positive Rate which is called Sensitivity are related to each other. Decision Tree covered the 0.62 area under the curve (AUC), Support Vector Machine covered the 0.68 area under the curve (AUC), Random Forest covers the 0.67 area under the curve (AUC) and Xtreme Gradient Boosting covered the 0.69 of AUC. The ROC curve was moderately fitted for a single class and another class fitted silly.

Local Classification

Local classification reveals that anxiety was divided into two categories anxious and not-anxious. These two categories were divided into four different categories as root nodes like minimal, mild, moderate, and severe levels. This Local Classification technique was used to classify the first layer of anxiety and then classify the second layer. Figure 5 displays the proposed local classification techniques for classifying anxiety.

(Fig. 5)

The different MLs were used to classify the first layer and second layers of anxiety. By employing the skcit-learn component for eleventh various operates, we determined the average score of accuracy, sensitivity, specificity, and precision of the DT, SVM, RF, and XGBoost algorithms to estimate anxiety disorders among undergraduate admission seekers. The results are shown below:

(Table 3)

Table 3

Models performance for the Local Classification
Indicators	DT	SVM	RF	XGBoost
Accuracy	0.763	0.818	0.811	0.819
Sensitivity	0.851	1.00	0.932	0.950
Specificity	0.317	0.00	0.224	0.500
F1-score	0.858	0.89	0.895	0.898
Positive Predictive Value	0.864	0.838	0.861	0.851

Table 3 displays the performance of the Local Classification techniques. This table shows that the better performance XGBoost which were accepted to classify the anxiety. Local Classification of the ML technique demonstrates superior accuracy (other indications) than the Flat Classification of the machine learning technology. To check the statement Local Classification was better than the Flat Classification, we had introduced the k-fold cross-validation based on accuracy.

(Fig. 6)

Figure 6 illustrates the estimated area under the curve (AUG) of DT, SVM, RF, and XGBoost models were implemented via programming language Python employing the scikit-learn bundle, with 70 percent of samples utilized as trained and 30 percent for testing and evaluation, and a randomly selected seed of 170158. Decision Tree covers the 0.63 area under the curve (AUC), Support Vector Machine covers the 0.65 area under the curve (AUC), Random Forest covered the 0.70 area under the curve (AUC) and Xtreme Gradient Boosting covers the 0.72 area under the curve (AUC). Local Classification of ML technique showed a better area under the curve than the flat classification of the machine learning technique.

K-Fold cross-validation of DT SVM RF and XGBT for Local Classification

Table 4 exhibits the outcome of cross-validation at a K-fold level for threefold, fivefold, and tenfold reiterates employing randomized seed 42 and the switch justification "Actual". The Xtreme Gradient Boosting model outperformed in threefold, fivefold, and tenfold cross-validations due to greater accuracy scores, i.e., 84%, 83%, and 84%, respectively as shown in Table 4.

Table 4

Result of accuracy based on K-Fold cross-validation of Local Classification model.
Models	Threefold	Fivefold	10-Fold
DT	0.75	0.76	0.76
SVM	0.82	0.81	0.81
RF	0.81	0.81	0.81
XGBoost	0.84	0.83	0.84

(Table 4)

Model Comparision

From the comparison, we got the random forest model was better than other methods using flat classification given in Fig. 7(A) and the xtreme gradient boosting model was better than other models using local classification given in Fig. 7(B). After that, we had to compare random forests using flat and xtreme gradient boosting using local classification.

(Fig. 7)

Figure 7 makes it clear that the best performer of local classification was better than the best performer of flat classification. The overall result showed that the Xtreme Gradient Boosting model using local classification was better than others given in Fig. 7(C).

Model to predict anxiety

Table 5 organizes the decision way for five students given data on their Gender, Family income, Drug addiction, Exercise, Pre-marital affairs, Study time, Practice religion, Blackmail, Family problem, Sickness, SSC result, and Mental problem for predicting their anxiety condition using the XGBT model.

Table 5

Prediction of undergraduate admission seekers’ anxiety using the fitted XGBT model.
Features												Predicted Outcome
Gender	Family income	Drug addiction	Exercise	Pre -Marital affairs	Study time	Practice religion	Black mail	Family problem	Sickness	SSC	Mental problem	Anxiety Layer 1	Anxiety Layer 2
Male	Lower-middle	Yes	No	Yes	3–4 hour	No	Yes	Yes	No	GPA 3	Yes	Anxious	Severe
Male	middle	No	Yes	Yes	5–6 hour	Yes	No	No	Yes	GPA 3–4	Yes	Anxious	Moderate
Female	middle	No	Yes	No	4–5 hour	No	Yes	Yes	Yes	GPA 3–4	Yes	Anxious	Moderate
Female	lower	No	Yes	No	5–6 hour	Yes	No	No	No	GPA 5	No	Not anxious	Minimal
Male	Lower-middle	Yes	No	Yes	3–4 hour	No	Yes	Yes	No	GPA 3–4	Yes	Anxious	Severe

(Table 5)

Students and admission candidates often face heightened vulnerability to anxiety and various other psychological wellness challenges, which can have detrimental consequences on both their physical wellness and academic achievements (Eisenberg et al., 2007). Worldwide, 60.8% of individuals experience moderate to significant severe levels of depression, 73% grapple with anxiety, and 62.4% contend with stress (Bayram & Bilgel, 2008). Previous research has found that anxiety affects as many as three-fifths of Bangladeshi university students which results support our study findings (Abu Hena Mostafa Alim et al., 2014, 2015; Mamun et al., 2021; Mamun, Hossain, et al., 2022; Rabby et al., 2023). There is a growing body of evidence indicating a rise in both the prevalence and severity of psychological health issues among students (Hunt & Eisenberg, 2010). Mental health issues can lead to significant psychological distress and heighten the likelihood of engaging in suicidal behaviors (Beiter et al., 2015). As a result, it is imperative to gain a comprehensive understanding and subsequently extend appropriate, impactful, and Easy to find treatment to this possibly endangered demographic (Beiter et al., 2015). Consultation emerges as the most commonly administered intervention, with substantial results observed in providers that use psychodynamic treatments, scheduled just short medication, and interdisciplinary therapist (Connell et al., 2008). Throughout campus counseling offerings in Australia, the UK, and the USA, there is a documented rise in students seeking help, often presenting with more severe issues (Eisenberg et al., 2007). However there doesn't seem to be unambiguous knowledge regarding university counseling facilities in Bangladesh, an appropriate number of research investigations have been conducted to try to decrease the high rate of psychological issues among undergraduates in the country (Arusha & Biswas, 2020a). The current research showed no substantial change in participants' anxiety levels concerning their relationship status, which predominantly agrees with the prior research by Marthoenis et al. (Marthoenis et al., 2018). Sleep duration and current living status are significantly associated with anxiety which was matched with our current study (Arusha & Biswas, 2020b). Females appeared more inclined than males to experience depression, anxiety, and stress symptoms, as well as a preference for admission to public universities, those who had a smaller monthly household income (25,000 BDT) were more likely to exhibit anxiety, stress, and depression-related symptoms reported by Rabby et al. (Rabby et al., 2023) which also coincided with our study findings.

The prior study for mental health prediction reported that in Sweden indicating non-significant superiority for the random forest model (AUC = 0.739, 95%CI 0.708–0.769) (Tate et al., 2020), in Pakistan where experimental evidence argues that RF outperforms its competitor classifiers (Ali et al., 2022), in India the most accurate one in Stacking technique based with an accuracy of prediction 81.75% (Vaishnavi et al., 2022). The previous studies in Bangladesh have reported that the RF model performed better in predicting stress and overall the lowest uncertainty estimates of those parameters, i.e., 81.9% of accuracy, 96.5% of precision, 67.5% of sensitivity, 88.5% of specificity, 87.2% of AUC, more than 88% of accuracy in all the 3, 5, and 10-folds cross-validation techniques among the university students (Rois et al., 2021) and Another study found that the random forest technique (accuracy = 0.893, precision = 0.715, sensitivity = 0.339, specificity = 0.979, F1-score = 0.460, AUC = 0.6613) accurately foresaw newborn mortality in comparison with additional machine learning (ML) methods taking into account individual and interaction influences (Rahman et al., 2021). our study, we conducted a comprehensive evaluation of various ML models, including Decision Trees (DT), Support Vector Machines (SVM), Random Forest (RF), and XGBoost, employing both flat classification in terms of traditional classification and local classification in term of hierarchical classification to predict anxiety. Specifically, The XGBoost model had an accuracy of 81.9%, an F1-score of 89.8%, sensitivity of 95.0%, specificity of 50.0%, and an AUC of 72%. Additionally, it maintained an accuracy of over 84% in all three, five, and ten-fold cross-validation techniques. These results collectively highlight the superior performance of the XGBoost model using local classification in accurately predicting anxiety levels in admission candidates. The XGBoost model meticulously assessed both individual factors and their interactions to predict anxiety levels among undergraduate admission candidates. This comprehensive approach is detailed in Table 5, illustrating how various factors contribute to the candidates' anxiety condition. In contrast, the Logistic Regression (LR) model encountered difficulties in estimation for the few significant predictors are independent, which can be attributed to shortcomings in the LR model's estimation process. Unlike XGBoost, the LR model necessitates strict adherence to underlying assumptions, particularly the requirement of significant associations between predictors and the outcome variable, as well as their independence to circumvent issues like multicollinearity. This analysis revealed that only a select few variables among the candidates were suitable for accurate estimation in the LR model. These variables demonstrated significant associations, as confirmed by the chi-squared test, not only with anxiety but also among themselves. This distinction was crucial in ensuring the LR model's efficacy in predicting anxiety levels. Therefore, to address the issue of multicollinearity, it is imperative to limit the inclusion of variables when estimating the LR model. Otherwise, the outcomes could potentially lead to misleading conclusions. Furthermore, the XGBoost model, distinguished by its freedom from assumptions during estimation, develops as a better and more trustworthy option. Given its outstanding performance and the absence of stringent assumptions, the XGBoost model stands out as the more authentic tool for predicting anxiety levels among undergraduate admission candidates in this study.

Limitation and strengthens

Furthermore, the use of a convenience sample is a key limitation of this study, which indicates that respondents in the study may not be representative the broader student population of Bangladesh. Another drawback of the study was its cross-sectional study, which could lead to variations over time. Despite the limits of the study, we believe our findings have some convincing advantages or strengths.

This study explores how to make use of several machine learning (ML) models in predicting the likelihood of undergraduate admission candidates' anxiety, such as DT, RF, and XGBoost, which don't necessitate any presumption and are highly straightforward (accessible) to integrate in any conventional software. Whereas the commonly utilized algorithm LR necessitates that all the fundamental requirements have been satisfied earlier than constructing the model, determinants are required to be autonomous from one another while maintaining a strong relationship with the outcome variable. As consequence, this frequently employed prognostic modeling is difficult to estimate well, and faulty projections may result in misleading data. The results we obtained can assist academics comprehend the limits of the renowned LR algorithm's assumptions-constrained feature. To properly run the LR model for this study, just a handful of variables from prospective employees must be employed for predicting anxiety; otherwise, the estimated model's outputs will be correct but less appropriate. Additionally, the model built on XGBoost used all thirteen relevant variables for predicting anxiety, adjusting for both individual and interaction effects. The XGBoost model, with its exceptionally high prediction accuracy, enhanced performance, and assumption-free features, will be more reliable and meaningful when utilized with a country-representative sample of significance to estimate the anxiety amongst undergraduate admission candidates.

Recommendation

Our research findings indicated a notable incidence of anxiety among undergraduate admission participants. Consequently, it is pivotal to Increase the amount of expertise among candidates and their parents about the factors that might indicate the development of anxiety. The framework for machine learning can be used to find and predict relevant prognostic markers for this anxiety problem with higher accuracy. There is a growing awareness of the need for a study to address the increased risk of psychological problems among undergraduate admission candidates in Bangladesh, but there is a severe lack of a psychological health system, counseling services, and policies in Bangladesh to help this preferably at-risk population.

This research adds to the evidence that there is an increased prevalence of anxiousness among university admission candidates. This psychological issue is extremely dangerous because it can have an impact on students' health, their admission exams, and confidence. Furthermore, the severity of this problem must be detected and understood to provide effective and suitable solutions for this vulnerable group. Machine Learning offers a powerful tool for identifying key prognostic factors and hierarchical classification techniques are other tools for enabling more precise prediction of psychometric issues. This enhanced accuracy empowers policymakers, stakeholders, and families to proactively address this critical crisis through refined policymaking, targeted psychological well-being, and the establishment of the highly efficient university and college consulting services. Colleges and universities may assign psychologists to service candidates whenever they want.

Conflict of interest

Funding

Data statement

Data will be provided upon reasonable request to the corresponding author

Authors contribution

Conceptualization: AA, MABS

Methodology: AA

Data curation: AA, AS, MABS

Formal analysis: AA, AS

Writing, editing and reviewing: All authors

Supervision: MAK, MABS

Abu Hena Mostafa Alim, S., Golam Rabbani, M., Karim, E., Syedul Islam Mullick, M., Al-Mamun, A., & Zillur Rahman Khan, M. (2015). Assessment of depression, anxiety and stress among first year MBBS students of a public medical college, Bangladesh.1. Bang J Psychiatry, 29(1), 23–29.
Abu Hena Mostafa Alim, S., Mahbub-E-Kibria, S., Jahurul ls, M., Zahir Uddin, M., Nessa, M., Abdul Wahab, M., & Mesbahul lslam, M. (2014). Translation of DASS 21 into Bangla and validation among medical students. Bang J Psychiatry, 28(2), 67–70.
Adewuya, A. O. (2006). Prevalence of major depressive disorder in Nigerian college students with alcohol-related problems. General Hospital Psychiatry, 28(2), 169–173.
Ali, M., Baqir, A., Sherazi, H. H. R., Hussain, A., Alshehri, A. H., & Imran, M. A. (2022). Machine Learning Based Psychotic Behaviors Prediction from Facebook Status Updates. Computers, Materials and Continua, 72(2), 2411–2427. https://doi.org/10.32604/cmc.2022.024704
Anderia, A. (2002). College student and depression pilot initiative. National Mental Health Assocation, 24(30), 4-5.
Arusha, A. R., & Biswas, R. K. (2020a). Prevalence of stress, anxiety and depression due to examination in Bangladeshi youths: A pilot study. Children and Youth Services Review, 116, 105254. https://doi.org/10.1016/j.childyouth.2020.105254
Arusha, A. R., & Biswas, R. K. (2020b). Prevalence of stress, anxiety and depression due to examination in Bangladeshi youths: A pilot study. Children and Youth Services Review, 116(May), 105254. https://doi.org/10.1016/j.childyouth.2020.105254
Bayram, N., & Bilgel, N. (2008). The prevalence and socio-demographic correlations of depression, anxiety and stress among a group of university students. Social Psychiatry and Psychiatric Epidemiology, 43(8), 667–672. https://doi.org/10.1007/s00127-008-0345-x
Beiter, R., Nash, R., McCrady, M., Rhoades, D., Linscomb, M., Clarahan, M., & Sammut, S. (2015). The prevalence and correlates of depression, anxiety, and stress in a sample of college students. Journal of Affective Disorders, 173(February), 90–96. https://doi.org/10.1016/j.jad.2014.10.054
Campbell, F., Blank, L., Cantrell, A., Baxter, S., Blackmore, C., Dixon, J., & Goyder, E. (2022). Factors that influence mental health of university and college students in the UK: a systematic review. BMC Public Health, 22(1), 1–22. https://doi.org/10.1186/s12889-022-13943-x
Connell, J., Barkham, M., & Mellor-Clark, J. (2008). The effectiveness of UK student counselling services: An analysis using the CORE System. British Journal of Guidance and Counselling, 36(1), 1–18. https://doi.org/10.1080/03069880701715655
Eisenberg, D., Gollust, S. E., Golberstein, E., & Hefner, J. L. (2007). Prevalence and Correlates of Depression, Anxiety, and Suicidality Among University Students. American Journal of Orthopsychiatry, 77(4), 534–542. https://doi.org/10.1037/0002-9432.77.4.534
Faye‐Dumanget, C., Carré, J., Le Borgne, M., & Boudoukha, P. A. H. (2017). French validation of the M aslach B urnout I nventory‐S tudent S urvey (MBI‐SS). Journal of Evaluation in Clinical Practice, 23(6), 1247–1251.
Hassan, T., Alam, M. M., Wahab, A., & Hawlader, M. D. (2020). Prevalence and associated factors of internet addiction among young adults in Bangladesh. Journal of the Egyptian Public Health Association, 95(1). https://doi.org/10.1186/s42506-019-0032-7
Hosen, I., Al-Mamun, F., & Mamun, M. A. (2021). Prevalence and risk factors of the symptoms of depression, anxiety, and stress during the COVID-19 pandemic in Bangladesh: a systematic review and meta-analysis. Global Mental Health, 8, e47.
Hunt, J., & Eisenberg, D. (2010). Mental Health Problems and Help-Seeking Behavior Among College Students. Journal of Adolescent Health, 46(1), 3–10. https://doi.org/10.1016/j.jadohealth.2009.08.008
Kulsoom, B., & Afsar, N. A. (2015). Stress, anxiety, and depression among medical students in a multiethnic setting. Neuropsychiatric Disease and Treatment, 11, 1713–1722. https://doi.org/10.2147/NDT.S83577
Mamun, M. A., Hossain, M. S., & Griffiths, M. D. (2022). Mental Health Problems and Associated Predictors Among Bangladeshi Students. International Journal of Mental Health and Addiction, 20(2), 657–671. https://doi.org/10.1007/s11469-019-00144-8
Mamun, M. A., Misti, J. M., Hosen, I., & Al Mamun, F. (2022). Suicidal behaviors and university entrance test‐related factors: a Bangladeshi exploratory study. Perspectives in Psychiatric Care, 58(1), 278–287.
Mamun, M. A., Rafi, M. A., Al Mamun, A. H. M. S., Hasan, M. Z., Akter, K., Hsan, K., & Griffiths, M. D. (2021). Prevalence and Psychiatric Risk Factors of Excessive Internet Use among Northern Bangladeshi Job-Seeking Graduate Students: A Pilot Study. International Journal of Mental Health and Addiction, 19(4), 908–918. https://doi.org/10.1007/s11469-019-00066-5
Marthoenis, Meutia, I., Fathiariani, L., & Sofyan, H. (2018). Prevalence of depression and anxiety among college students living in a disaster-prone region. Alexandria Journal of Medicine, 54(4), 337–340. https://doi.org/10.1016/j.ajme.2018.07.002
Mateen, B. A., Liley, J., Denniston, A. K., Holmes, C. C., & Vollmer, S. J. (2020). Improving the quality of machine learning in health applications and clinical research. Nature Machine Intelligence, 2(10), 554–556. https://doi.org/10.1038/s42256-020-00239-1
Nordin, N. M., Talib, M. A., & Yaacob, S. N. (2009). Personality, loneliness and mental health among undergraduates at Malaysian Universities. European Journal of Scientific Research, 36(2), 285–298.
Over 95 % of students clear HSC , equivalent exams. (2022). Dhaka Tribune. https://www.dhakatribune.com/bangladesh/education/263708/over-95%25-of-students-clear-hsc-equivalent-exams
Ovuga, E., Boardman, J., & Wasserman, D. (2006). Undergraduate student mental health at Makerere University, Uganda. World Psychiatry : Official Journal of the World Psychiatric Association (WPA), 5(1), 51–52.
Protikuzzaman, M., Baowaly, M. K., Devnath, M. K., & Singh, B. C. (2020). Predicting undergraduate admission: a case study in Bangabandhu Sheikh Mujibur Rahman Science and Technology University, Bangladesh. International Journal of Advanced Computer Science and Applications, 11(12), 138–145. https://doi.org/10.14569/IJACSA.2020.0111217
Rabby, M. R. A., Islam, M. S., Orthy, M. T., Jami, A. T., & Hasan, M. T. (2023). Depression symptoms, anxiety, and stress among undergraduate entrance admission seeking students in Bangladesh: a cross-sectional study. Frontiers in Public Health, 11. https://doi.org/10.3389/fpubh.2023.1136557
Rahman, A., Hossain, Z., Kabir, E., & Rois, R. (2021). Machine learning algorithm for analysing infant mortality in Bangladesh. International Conference on Health Information Science, 205–219.
Roberts, M., Driggs, D., Thorpe, M., Gilbey, J., Yeung, M., Ursprung, S., Aviles-Rivero, A. I., Etmann, C., McCague, C., Beer, L., Weir-McCall, J. R., Teng, Z., Gkrania-Klotsas, E., Ruggiero, A., Korhonen, A., Jefferson, E., Ako, E., Langs, G., Gozaliasl, G., … Schönlieb, C. B. (2021). Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans. Nature Machine Intelligence, 3(3), 199–217. https://doi.org/10.1038/s42256-021-00307-0
Saleem, S., & Mahmood, Z. (2013). Mental Health Problems in University Students: A Prevalence Study. FWU Journal of Social Sciences, 7(2), 124–130.
Seim, R. W., & Spates, C. R. (2009). The prevalence and comorbidity of specific phobias in college students and their interest in receiving treatment. Journal of College Student Psychotherapy, 24(1), 49–58. https://doi.org/10.1080/87568220903400302
Siddik, M. A. B., Ali, A., Miah, S., Hasan, M., Ahmed, M., & Sunna, T. C. (2024). Psychological disorders among college going students: A post Covid-19 insight from Bangladesh. Journal of Affective Disorders Reports, 15, 100686.
Siddik, M. A. B., Hasan, M. N., Mahmud, A., Munmun, M. S., Milad, M. H., Ali, A., Ahmed, Z., & Uddin, M. J. (2023). Prevalence of depression and its associated factors among undergraduate admission candidates in Bangladesh: A nation-wide cross-sectional study. PLoS One, 18(11), e0295143.
Sweatt,S.K, Gower, B.A, Chieh, A.Y, Liu, Y, Li, L. (2016). 乳鼠心肌提取 HHS Public Access. Physiology & Behavior, 176(1), 139–148. https://doi.org/10.1016/j.jad.2017.07.044.MENTAL
Tate, A. E., McCabe, R. C., Larsson, H., Lundström, S., Lichtenstein, P., & Kuja-Halkola, R. (2020). Predicting mental health problems in adolescence using machine learning techniques. PLoS ONE, 15(4), 1–13. https://doi.org/10.1371/journal.pone.0230389
ul Haq, M. A., Dar, I. S., Aslam, M., & Mahmood, Q. K. (2018). Psychometric study of depression, anxiety and stress among university students. Journal of Public Health (Germany), 26(2), 211–217. https://doi.org/10.1007/s10389-017-0856-6
UN. (n.d.). The Sustainable Development Goals in Bangladesh. United Nations Bangladesh. https://bangladesh.un.org/en/sdgs/3
Vaishnavi, K., Kamath, U. N., Rao, B. A., & Reddy, N. V. S. (2022). Predicting Mental Health Illness using Machine Learning Algorithms. Journal of Physics: Conference Series, 2161(1). https://doi.org/10.1088/1742-6596/2161/1/012021
Verger, P., Guagliardo, V., Gilbert, F., Rouillon, F., & Kovess-Masfety, V. (2010). Psychiatric disorders in students in six French universities: 12-month prevalence, comorbidity, impairment and help-seeking. Social Psychiatry and Psychiatric Epidemiology, 45, 189–199.
Wald, A., Muennig, P. A., O’Connell, K. A., & Garber, C. E. (2014). Associations between healthy lifestyle behaviors and academic performance in U.S. undergraduates: A secondary analysis of the american college health association’s national college health assessment ii. American Journal of Health Promotion, 28(5), 298–305. https://doi.org/10.4278/ajhp.120518-QUAN-265

No competing interests reported.

Download PDF

Editorial decision: Revision requested
28 Aug, 2024
Reviews received at journal
12 Aug, 2024
Reviewers agreed at journal
05 Aug, 2024
Reviewers agreed at journal
05 Aug, 2024
Reviewers invited by journal
04 Aug, 2024
Editor assigned by journal
02 Aug, 2024
Editor invited by journal
23 Jul, 2024
Submission checks completed at journal
23 Jul, 2024
First submitted to journal
09 Jul, 2024

You are reading this latest preprint version

Applications of Hierarchical Classification Techniques for Classifying Anxiety Among Undergraduate Admission Candidates in Bangladesh

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Introduction

Materials and methods

Ethical clearance

Study design and data collection procedure

Sample size and sampling technique

Criteria for inclusion and exclusion

Measures

Socio-demographic determinants

General Anxiety Disorder

Statistical Analyses

Results

Selection of features

(Fig. 1)

Descriptive analysis and cross-tabulation of the selected feature

(Fig. 2)

Flat Classification Technique

Local Classification

K-Fold cross-validation of DT SVM RF and XGBT for Local Classification

Model Comparision

(Fig. 7)

Model to predict anxiety

Discussion

Limitation and strengthens

Recommendation

Conclusion

Declarations

References

Additional Declarations

Status:

Version 1