TAGRA 2025

Demand fluctuations and increasing product variety make production planning a significant challenge in the textile industry. In environments where parallel machines operate with varying capacities and sequence-dependent setup requirements, accurate demand forecasting becomes essential for establishing realistic schedules and improving resource utilization. This study focuses on developing a reliable machine learning–based demand forecasting model to support these planning decisions.
The research uses monthly sales data obtained from a textile manufacturer in Türkiye. Several regression algorithms, including Random Forest, Gradient Boosting and CatBoost, were examined to determine the most effective method for predicting future demand. The dataset was enriched through feature engineering by incorporating seasonal indicators, lagged demand variables and trend components. All models were evaluated using repeated cross-validation to ensure stable performance and to prevent overfitting. Model accuracy was assessed through widely used metrics such as RMSE, MAE and R².
The results indicate that the CatBoost algorithm provides the highest prediction accuracy among the tested models. Its superior performance contributes to clearer demand visibility and supports planners in making more consistent capacity allocations. Improved forecasts also help reduce bottlenecks related to setup times and enable more balanced machine loading.
Overall, the proposed model offers a practical, data-driven approach for textile manufacturers to improve production planning. It also lays the groundwork for a broader study, where the demand forecasts will be used as inputs in developing an optimization model for parallel machine scheduling with sequence-dependent setup times. By integrating accurate forecasts into scheduling decisions, this approach helps reduce bottlenecks, ensures more balanced machine utilization, and strengthens overall planning reliability.