ADVANCED STRESS LEVEL FORECASTING USING QUANTILEOPTIMIZED REGRESSION FOREST MODELS

Abstract

The accurate prediction of human stress levels has become increasingly important in intelligent healthcare, workplace monitoring, and mental well-being assessment systems. Traditional regression models often struggle to capture the asymmetric distribution and uncertainty inherent in physiological and behavioral stress data. This study proposes an enhanced stress forecasting framework using a Quantile-Optimized Regression Forest (QORF) model refined with a pinball loss function to improve prediction accuracy across varying stress intensities. The method integrates physiological indicators, contextual features, and behavioral markers to generate a comprehensive stress estimation profile. By leveraging quantile learning, the system effectively models uncertainty and provides more reliable upper- and lower-bound predictions compared to conventional mean-based approaches. Experimental evaluations conducted on benchmark stress datasets demonstrate notable improvements in prediction robustness, reduced error margins, and superior handling of skewed data distributions. The findings confirm that the proposed QORF–Pinball Loss framework is a powerful and efficient solution for real-time stress level forecasting in intelligent monitoring applications