Machine learning‐based prediction of functional groups and colour strength in reactive dyed protease‐treated polyester

Abstract Polyester, due to its hydrophobic nature and low chemical reactivity, requires a high‐temperature, high‐pressure (HTHP) dyeing process that is both energy‐ and resource‐intensive. This challenge becomes even more pronounced when dealing with polyester‐cotton blends, which demand longer dyeing durations, introduce additional operational complexities and increase wastewater loads. In response to these limitations, this research explores the protease enzyme treatment as a method to improve the reactivity and dyeability of polyester. It incorporates machine learning (ML), specifically XGBoost, to predict key surface and dyeing parameters. Experimental results demonstrate that enzymatic hydrolysis successfully introduces OH and COOH groups, verified by FTIR. Importantly, the treatment does not significantly compromise the fabric's mechanical or thermal properties. The XGBoost model demonstrated exceptional prediction accuracy ( R 2 > 0.98) for both functional groups and surface color strength of all dyed specimens. Shapley Additive Explanations (SHAP) analysis identified enzyme concentration as the most influential factor, followed by pH, treatment duration and temperature. Among the dyes, CI Reactive Blue 4 achieved the most reliable predictive performance. Nevertheless, all models demonstrated operational reliability with MAPE below 3%, suggesting robust potential for sustainable process control in polyester dyeing. The experiment‐ML‐based methodology offers a practical pathway towards environment friendly textile processing.