Efficient production capacity planning is essential for industries facing uncertain demand, such as semiconductor manufacturing. This thesis explores machine learning models to improve demand forecasting and optimize production capacity, focusing on the ASML Field Service Hub. Models including XGBoost, Random Forest, and ARIMA are evaluated to predict demand across multiple product flows under uncertainty. These forecasts are integrated into an optimization model to minimize costs related to overcapacity, undercapacity, and labor fluctuations.
The proposed framework demonstrates substantial improvements over ASML’s current benchmarks, reducing forecast errors by up to 37.4% and simulated costs by 48.3%. Additionally, robustness testing was conducted using jitter tests, ensuring the model’s stability in the face of noisy data. The explainability and feature importance of the machine learning models were investigated using SHAP values. The study’s results highlight the model’s adaptability, offering a valuable tool for improving capacity planning in various industries dealing with demand volatility.