Explainable Depression Detection From Text Using Machine Learning and Transformer-Based Models

The increased likelihood of depression among university students is a concern that needs early diagnosis to ensure early intervention measures are taken. Recent breakthroughs in natural language processing (NLP) technology have seen the capability to automatically determine depression from text data become possible. This article describes a comparative analysis conducted by combining traditional ML models and transformer models for depression analysis. The dataset for this experiment is derived from 7,731 cleaned pieces of text with an unbalanced number of classes representing each set of data. The experiment reveals that XGBoost outperforms other traditional models with a precision rate of almost 99 % and a recall rate of 98.6 % for the depression class. The transformer based DistilBERT model is also shown to perform well, providing excellent context for the analysis of each text piece. Explainable AI is introduced to improve the analysis models by providing explanations for the results derived.