Aprendizado de máquina para a predição da saúde mental e qualidade de vida de estudantes universitários

Abstract

In recent years, there has been a significant increase in the number of studies focused on Mental Health (MH) worldwide. Students, from elementary school to university, have gained notoriety due to the rise in anxiety and depression rates. In the task of identifying the causes of these elevated numbers, sociodemographic indicators and data related to mental health and quality of life are of utmost importance. Therefore, this study aims to evaluate the effectiveness of using machine learning techniques to predict the levels of MH and Quality of Life (QoL) of students at the University of Vale do Acaraú (UVA) and to identify the main causes of anxiety and depression experienced by them. The dataset used consists of responses from 880 students to three questionnaires: one related to sociodemographic, academic, and clinical data, the Mental Health Inventory (MHI), and the WHOQOL-Bref. Data processing was based on the concepts and processes of Data Mining (DM) and Knowledge Discovery in Databases (KDD), and the trained algorithms were: Multilayer Perceptron (MLP), Support Vector Machine (SVM), Random Forest (RF), and Adaptive Boosting (AdaBoost). After hyperparameter tuning, done via grid search, and feature selection performed with the Sequential Forward Floating Selection (SFFS) and Sequential Backward Floating Selection (SBFS) algorithms, the results achieved an accuracy of 78.44% for the prediction of MH and 77.40% for the prediction of QoL, both using RF as the model. Additionally, the importance of the attributes was evaluated by calculating information gain and stepwise regression, with the most important attributes for MH being the scores of the WHOQOL-Bref domains, and for QoL, the scores of the MHI domains.