A machine learning model for the identification of depressive symptoms among university students in Bangladesh

BACKGROUND: The present study aimed to identify a machine learning (ML) model exploring depressive symptoms among university students in Bangladesh (i.e., to evaluate which type of algorithm best identifies depression based on participant characteristics only). METHODS: A total of 346 randomly selected students studied at Khulna University (Bangladesh) were considered for analysis. Depressive symptoms were assessed utilizing the Centre for Epidemiologic Studies Depression Scale (CESDS). RESULTS: The data were split into two separate datasets, with the first dataset for training (75%) and the second dataset for testing (25%). All five well-known ML algorithms were applied to train the selected models. The predictive performances of these algorithms were compared based on the performance parameters such as accuracy, precision, sensitivity, F1-score, and area under the curve (AUC). Among the classifiers, the gradient boosting machine (GBM) algorithm proved to be the best, with a maximum accuracy of 89%, maximum precision of 88%, a higher sensitivity of 89%, and a maximum F1 score of 94%. Additionally, the best discriminative ability also the GBM classification (AUC=0.89). GBM algorithm best identified depressive symptoms among Bangladeshi university students compared to the other ML algorithms applied in the study. CONCLUSIONS: Psychologists and counsellors may utilize the GBM algorithms to identify depression among students so that appropriate steps can be taken to reduce the burden of the depressive symptoms among students.