Enhancing Decomposition and Holt-Winters Weekly Forecasting of PM2.5 Concentrations in Thailand’s Eight Northern Provinces Using the Cuckoo Search Algorithm

Abstract

This research aims to introduce hybrid models that integrate the Cuckoo Search Algorithm with Holt-Winters (CS-HW) and Decomposition (CS-D) for time series forecasting of weekly PM2.5 concentrations in Thailand’s eight northern provinces. The study consists of two phases: the training dataset phase and the testing dataset phase. During the training dataset phase, the Cuckoo Search (CS) algorithm demonstrates effective parameter optimization capabilities, seamlessly integrating with Holt-Winters and decomposition models. This integration results in lower Root Mean Square Error (RMSE) values compared to classical approaches, including Grid Search for Holt-Winters (Classic-HW) and Classical Decomposition (Classic-D). In the testing dataset phase, key performance metrics such as RMSE, Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) are utilized. The results indicate that the CS-HW and CS-D models outperform other methods in weekly forecasting of PM2.5 concentrations across several provinces, including Chiang Mai, Chiang Rai, Lamphun, Lampang, Mae Hong Son, and Phayao. Notably, the Box-Jenkins model outperformed other methods in Nan, while in Phrae, the Long Short-Term Memory (LSTM) model demonstrates other forecasting performance.