Textile industry, particularly wet processing and thread dyeing industry, produces several waste flows that creates serious environmental and operational problems. This research aims to develop a combined data-driven model for predictive waste classification and reverse logistics optimization in a circular supply chain system. The study is focused on integrating daily operational data of two years from a thread dyeing company,characterized by features, such as production volume, dye batch count, chemical and dye usage, water and energy consumption, machine utilization and waste generation, to train Random Forest machine learning model for predicting waste quantities. With the outcome of the waste prediction, A Mixed Integer Linear Programming Problem (MILPP) for reverse logistics strategies was implemented to optimize waste collection, recycling, treatment, and disposal decisions across the supply chain. The developed framework will enable the manufacturers to plan their waste management more accurately, save operating costs, and improve recovery of valuable resources within textile manufacturing operations. The findings demonstrated the effectiveness of the model, as a scalable, data-driven decision-support tool for circular economy, as well as other resource-intensive manufacturing industries.