Deep Learning for Early Detection of Depression in University Students: A Mental Health Perspective

Abstract

Nowadays, depression has spread at an alarming rate among students. There are many reasons for this such as lack of sleep, academic decline, loneliness, limited mental support, personal, family or financial problems, etc. In this thesis, I used 1977 datasets where there were 15 universities. Out of which 9 were public and 6 were private. The data was surveyed using PHQ-9 method. It has 6 levels. My proposed method is ANN- artificial neural network. Through which I got 97.98% accuracy.my dataset is a tabular form such as mixed with demographic, academic, phycological. Poor concentration, lack of sleep, and feelings of low self-worth were more influential predictors of depression than demographic and academic information. For better understanding, i used explainable ai (LIME), which full form is Local Interpretable Model Agnostic Explanation. with the help of this method, we can easily understand the reason of the depression which is causing more effectively. I also preprocess the dataset for cleaning the data and understanding for human language to machine language like labeling hot encoding etc. I also used PCA-IG feature selection method for feature selection and changed the hyperparameter of each model for knowing which one is more suitable and will give us more accuracy. The results can help shape targeted mental health programs, guide policy decisions, and inspire future research that tracks students over time and involves more universities—helping address mental health needs in settings where resources are limited.