The physical, social, and mental conditions of machine learning in student health evaluation

Background

This study aims to assess how well several machine learning (ML) algorithms predict the physical, social, and mental health condition of university students.

Objectives

The physical health measurements used in the study include BMI (Body Mass Index), %BF (percentage of Body Fat), BSC (Blood Serum Cholesterol), SBP (Systolic Blood Pressure), and DBP (Diastolic Blood Pressure).

Methods

The mental health evaluation relied on the following methods: PHQ-9 (Patient Health Questionnaire-9), ISI (Insomnia Severity Index), GAD-7 (Generalized Anxiety Disorder Scale), and SBQ-R (Suicidal Behaviors Questionnaire-Revised). The study assessed KEYES, the comprehensive social health indicator. The study uses a famous methodology for training and testing four well-known ML algorithms, namely the K-nearest neighbors algorithm, decision trees, Naïve Bayes, and the random forest algorithm.

Results and Conclusions

The recall value of the RF algorithm is higher by 2.0%, 4.15%, and 11.25%, respectively. The F-score value of the RF algorithm is also the highest. The differences amount to 4.56% (Naïve Bayes), 2.50% (DT), and 11.20% (K-NN). Accuracy, Precision, Recall, and F-score were used to assess the researched ML algorithms' prediction ability. With a 99.40% prediction accuracy, a 97.60% precision, a 99.30% recall, and an F-score value of 98.70%, the Random Forest method performed the best. ML algorithms can serve as tools for the prediction of physical, mental, and social health state of patients, including students, but they have a rather narrow scope of application and do not cover all aspects of health.