Optimizing Combined Holt–Winters and Decomposition Forecasting Models with the Whale Optimization Algorithm for Monthly Maximum of Daily 24-hour Average PM2.5 Concentrations in the Bangkok Metropolitan Region

Abstract

This study proposes a novel hybrid forecasting model, Combined Holt–Winters and Decomposition optimized by the Whale Optimization Algorithm (CHD–WOA), to predict the monthly maximum of daily 24-hour average PM2.5 concentrations in the Bangkok Metropolitan Region. The model integrates decomposition and Holt–Winters exponential smoothing forecasts through weighted averaging, with 18 parameters simultaneously optimized using the Whale Optimization Algorithm. Ten methods are compared, including WOA-optimized variants, classical combination models, and benchmark approaches such as Box–Jenkins and Long Short Term Memory (LSTM). Performance is evaluated using RMSE for in-sample fit and MAPE for out-of-sample accuracy. The proposed CHD–WOA model achieves the lowest MAPE at four of eight stations and consistently outperforms benchmarks. Forecasts for 2025 accurately capture seasonal pollution peaks during the dry season, demonstrating the robustness and effectiveness of combining classical statistical techniques with metaheuristic optimization for complex urban air quality forecasting.