Analyze Factors Affecting Thread Consumption in a Garment and Develop a Machine Learning-based Prediction Model

The garment manufacturing industry faces challenges related to excess thread stock, leading to increased write-off expenses and environmental impact. This research utilizes Machine Learning (ML) techniques to predict sewing thread consumption in producing full underwear briefs, considering key factors such as garment style, fabric/seam thickness, stitch length, seam type, and estimated wastage. Multiple ML models, including Linear Regression, Ensemble Models (Random Forest, Gradient Boosting), and Artificial Neural Networks (ANN), were employed to predict thread consumption for two thread types (TKT 120 and TKT 160). These models are integrated into a user-friendly Streamlit interface to make real-time predictions based on user input, improving inventory management decision-making. ANN models outperformed others, with a validation dataset MAPE of $\mathbf{1. 2 1 \%}$ for TKT 120 and $1.99 \%$ for TKT 160. This study shows how ML optimises thread usage, reduces inventory costs, and supports garment manufacturing sustainability. The ANN model performs well, but refinement and feature exploration could improve prediction accuracy and generalization. The research emphasises ML and usercentric tools for industry resource management efficiency.