Anongphorn Janboonpeng and Chompoonoot Kasemset
Published in Data Science and Engineering (DSE) Record 2025 Vol. 6 No. 1 pp. 505-516
Abstract
Customer demand volatility in rapidly changing market environments poses significant challenges to production planning, particularly for short life-cycle products that exhibit pronounced seasonal patterns. Exploratory Data Analysis (EDA) conducted in this study revealed clear annual cycles and distinct seasonal peak periods in customer demand, highlighting the need for forecasting methods capable of effectively and reliably capturing such seasonality. Therefore, this research aims to develop a Forecasting Method Evaluation Framework to guide the selection of appropriate forecasting models for production planning under uncertain demand conditions. The proposed framework consists of five key components: (1) data preparation and seasonal feature engineering informed by EDA findings, (2) development of statistical and deep learning forecasting models, including SARIMA, Holt-Winters, LSTM, GRU, and a hybrid SARIMA–LSTM model, (3) performance evaluation using MAE, RMSE, MAPE, and R², and (4) integration of forecast outputs into production planning processes to effectively accommodate demand variability. The framework supports improved production stability by reducing the frequency of production plan adjustments and enhances operational readiness across manpower, machinery, materials, and production methods (Man–Machine–Material–Method), ensuring alignment with future demand conditions.