|
Short-term demand forecasting for production planning: a case study of optical product |
|---|---|
| รหัสดีโอไอ | |
| Title | Short-term demand forecasting for production planning: a case study of optical product |
| Creator | Nilrat Yodngern |
| Contributor | Pham Duc Tai, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | Demand forecasting, Time-series analysis, Eyewear Manufacturing, Statistical forecasting models |
| Abstract | This study addresses short-term demand forecasting in eyewear medical device manufacturing, where limited historical data and volatile daily demand complicate accurate prediction. Daily lens demand data over a two-year period were analyzed using traditional time-series models, including AR, MA, ARMA, ARIMA, simple exponential smoothing, and double exponential smoothing, and compared with an Artificial Neural Network (ANN). Among the traditional models, ARIMA (3,1,4) achieved the best final test performance, with RMSE of 1,992.65 and MAPE of 24.90%. The selected ANN used engineered calendar, holiday, lag, and rolling statistical features, with 16 hidden units, ReLU activation, Adam optimizer, learning rate of 0.0085, and batch size of 32. After retraining the combined training and validation data, the ANN achieved MAPE of 10.18% on the final 7-day test horizon. The results indicate that ANN improved short-term forecasting accuracy by capturing nonlinear demand patterns, although further validation across multiple 7-day test windows is recommended to confirm model stability. |
