Data-Driven Decision Support System for Managing Discrepancies in Textile Ordering Operations

Subcontracted dyeing in the textile industry frequently causes discrepancies between the quantities of fabric sent and received, resulting in stock imbalances and scheduling inefficiencies. This study presents a machine learning (ML) based decision support system, developed using real industrial data, to predict fabric losses and width, thereby reducing uncertainty. The framework integrates two regression models (for fabric loss percentage and final width), a classification model for categorical loss prediction, and an unsupervised clustering model to identify production patterns. Implemented with XGBoost and validated through 10-fold cross-validation, the models achieved high accuracy and interpretability. The proposed approach enables planners to anticipate losses, optimize subcontracting quantities, and improve inventory reliability. Results show a significant potential to reduce stock shortages and enhance subcontracting decisions, demonstrating the practical impact of AI-based decision support in textile manufacturing.