Machine Learning Based Depression Prediction: Comparative Analysis of Models and Stacked Ensemble Approach

Depression is a multifaceted mental illness that has a significant impact on a person's thoughts, feelings, and general state of well-being. This presents a serious problem for students which not only affects their academic achievement but also affects their mental health. A panoptic analysis of machine learning techniques for depression prediction is presented in this paper. Through a Google survey targeting university students, we collected a dataset of 550 examples covering a wide range of relevant features. Five widely used machine learning models were employed after efficient data preprocessing and feature selection: Logistic Regression, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision trees, and Neural networks. After complete evaluation of all pre-trained models among them Neural network outperforms others with the accuracy of 94.55%. Based on this success, an improved combination method including SVM, KNN and Logistic Regression with a Neural network as a meta-model was developed. This stacking ensemble learning method provided an outstanding accuracy of 96.36%, surpassing the performance of individual models. The proposed model pushes the boundaries of machine learning in mental health analytics, demonstrating tremendous potential as a powerful solution for depression prediction.