An Ensemble Learning Model for Accurate Mental Stress State Prediction

The tremendous use of technology and social media and also tremendous work from working place so that people suffer from several mental disorders like anxiety, and stress become critical global concerns, affecting individuals’ quality of life and overall productivity. This study proposes an intelligent system that integrates NLP and ML to predict the mental state issues by analyzing linguistic and behavioral patterns from textual data on social media. This paper proposed the ML based ensemble model that involves selecting the feature such as emotional tone, sentiment polarity, and word frequency, followed by text preprocessing techniques like tokenization, stop-word removal, and lemmatization. Various baseline ML models were proposed and trained and evaluated to identify individuals at potential risk of low, medium and high. XGBoost obtained accuracy of 0.9636, F1-score of 0.9594, and MCC of 0.9409. The Light GBM obtained an accuracy of 0.9573 and F1-score of 0.9526. RF obtained accuracy of 0.9281 and an MCC of 0.8820. The ensemble model shows the best overall results to obtained accuracy of 0.9842, MCC of 0.9744 and Cohen’s kappa of 0.9743. The findings show the strong capability of NLP-based ML approaches in detection of mental health conditions.