Stress Level Prediction Using Pinball Loss Function Based Quantile Regression Forest Approach

Depression, anxiety, and stress stand as the prevailing mental health challenges, affecting individuals worldwide with profound consequences.Identification and determination of these mental illnesses are important for preventing their harmful effects.However, overfitting concerns arise due to the limited sample size utilized in Machine Learning (ML) techniques to obtain more robust and accurate models.Hence, this research proposes Pinball Loss Function based Quantile Regression Forest (PLF-QRF) approach for the prediction of stress level.The PLF-based QRF enhances this model, as well as minimizes outliers by reducing the sum of weighted absolute errors over quantiles.Initially, DASS-42 evaluation method is used to validate the model's effectiveness which is pre-processed by using min-max normalization for handling the class imbalance issues by employing Synthetic Minority Over-Sampling Technique (SMOTE).Then, feature selection is performed utilizing Sparse Principal Component Analysis (SPCA).Eventually, the PLF-QRF technique is used for the prediction of stress levels.The PLF-QRF attains better accuracy of 99.51% based on DASS-42 when compared to the existing methods like stacking ML and ensemble ML.