Accurate assessment of heat stress is essential for protecting outdoor workers in hot–humid environments, yet direct field measurement of the Wet-Bulb Globe Temperature (WBGT) remains resource intensive and impractical for continuous monitoring. Meteorology-based predictive models offer a scalable alternative but often require a large number of input variables, limiting real-world implementation. This study aimed to develop and validate a simplified WBGT prediction model that retains the predictive accuracy of a full model while substantially reducing input complexity, thereby improving usability in practical occupational settings. A dataset of 450 observations was constructed by pairing field-measured WBGT with 25 meteorological predictors obtained from an automated weather source. Data preprocessing included removal of missing values, near-zero variance filtering, unit standardization, numerical encoding of area and time variables, and computation of wet-bulb temperature. The Extreme Gradient Boosting (XGBoost) algorithm was selected as the primary modeling technique due to its superior performance compared with linear regression, decision tree, and random forest models under 5-fold cross-validation. Model training was conducted in R (version 4.3.2) using fixed hyperparameters and a reproducible seed. To optimize the model for practical use, Recursive Feature Elimination (RFE) coupled with XGBoost was applied to identify a minimal subset of predictors. This process reduced the number of variables from 25 to 8 (a 68% reduction). The selected variables: DiffuseRadiation, SunshineDuration, CloudCoverHigh, Area, Temperature_2m, Time, WetBulb_2m, and DewPoint_2m represented essential physical determinants of WBGT, including thermal, humidity, radiation, and spatiotemporal components. Model performance was evaluated using MAE, RMSE, and R2. The Reduced Model achieved MAE = 0.6464°C, RMSE = 0.8738°C, and R2 = 0.8620, demonstrating performance comparable to the Full Model (MAE = 0.6628°C, RMSE = 0.8980°C, R2 = 0.8568). A predefined practical equivalence margin of ±0.10°C was used to determine operational similarity. Non-inferiority and Two One-Sided Tests (TOST) confirmed that the Reduced Model was statistically equivalent to the Full Model for both MAE and RMSE. The simplified model reduces data collection burden, minimizes user errors, and supports field deployment while maintaining high predictive accuracy. These findings highlight the model's potential as an efficient, user-friendly tool for occupational heat stress management.