Student Mental Health Risk Classification Using Random Forest and BERTopic: A Tabular and Text Analysis Approach

Abstract

Student mental health has become a critical issue affecting academic performance and quality of life. Many previous studies have only used numerical data and have not explored textual data that reflects students' subjective conditions. This study develops a classification model using Random Forest on tabular data to predict the risk of student mental health issues. The dataset consists of 500 students with a training-to-testing data split of 80:20. The model uses features such as academic performance, sleep patterns, and psychological scores. The results show that the model achieves a macro F1 Score of 0.90 despite class imbalance. The most influential variables are depression scores, stress levels, and anxiety. In addition, the BERTopic method was used separately to analyze text data and identify thematic patterns from student responses. However, not all topics generated were directly related to mental health. The model also has limitations in detecting minority classes due to data imbalance. This study provides a classification approach supported by text analysis to comprehensively understand students' mental health conditions.