Machine Learning-Based Forecasting of Occupational Heat Stress in Manual Outdoor Work Systems

Introduction
Occupational heat stress threatens safety and productivity in outdoor manual labor, particularly in tropical climates where high humidity amplifies thermal strain. Reliable forecasting of these conditions is essential for preventive planning and work–rest scheduling. However, traditional predictive approaches that rely on linear regression often struggle to capture the complex, non-linear interactions between microclimatic variables and human thermal exposure. This study investigates the potential of machine learning (ML) techniques to improve the forecasting accuracy of occupational heat stress indicators.

Methods
The study revisits a 2020 field dataset containing microclimatic observations (ambient temperature, relative humidity, and surface characteristics) from outdoor work settings. A machine learning-based predictive framework was developed to estimate Physiological Equivalent Temperature (PET). Three algorithms, Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Regression (GBR), were implemented to these capture non-linear relationships. Model performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²).

Results
The preliminary modeling results indicate that ensemble-based approaches, particularly RF and GBR, outperform conventional regression models when predicting periods of elevated afternoon heat stress. The models demonstrate an improved capability to identify the extreme thermal conditions associated with increased operational risk.

Conclusions
The findings demonstrate the potential of machine learning models to forecast occupational heat stress. While it is constrained by the localized dataset and requires future real-time field validation, the framework provides a practical basis for integrating data-driven forecasting into heat-mitigation strategies and occupational safety planning.