Depression Analysis Based on Machine Learning and Model Visualization

Abstract

Depression is among the most prevalent mental health issues in the world. Traditional diagnostic methods rely on clinical interviews and scoring scales, which are subjective and biased. In this paper, the adopted model is based on a massive dataset with more than 27,901 student samples to conduct experiments, where this study compares three machine learning algorithms, including Random Forest (RF), Support Vector Machine (SVM) and K-Nearest Neighbor (K-NN). The methods on SVM, RF, and K-NN preprocessed, i. e., excluding outliers through the Inter Quartile Range (IQR) method, feature encoding, and Z-score normalization procedure. The experimental results shows that all models achieved good performance, with SVM having the highest accuracy followed by RF then K-NN. This study also analyzed the confusion matrix in detail. It indicates that SVM performs well in positive class recognition, while random forest excels in negative class recognition. The research results indicate that machine learning models can capture the nonlinear relationships in depression data. They can also handle complex interactions between different factors. This is something traditional linear regression methods cannot do well. This study offers important new insight for automatic depression screening. It also contributes to the development of objective diagnostic tools in mental health care.