Predictive modeling and chemometric optimization in natural dyeing systems: An analytical case study of wool dyeing with Melissa officinalis l

Abstract

This study introduces a robust chemometric framework that integrates predictive modeling and multi-criteria decision-making for the analytical assessment of natural dye systems. Using Melissa officinalis L. as a model extract, we demonstrate an eco-conscious dyeing application on wool yarns evaluated through objective colorimetric parameters. A total of 40 treatment combinations involving bio- and metal mordants were assessed. The Weighted Aggregated Sum Product Assessment (WASPAS) method was used to rank treatments based on L*, a*, and b* values, identifying the Cu-GA combination as optimal with a composite score of 1.71. To model dyeing behavior, we implemented a feedforward Artificial Neural Network (ANN) trained on 3720 K/S data points across treatment and wavelength conditions. The ANN achieved high predictive accuracy (R2 = 94.13-95.28; MSE = 1.37-1.95) using Levenberg-Marquardt backpropagation. This model enabled the interpolation of unmeasured color strength values, enhancing reproducibility and reducing experimental load. UV protection was also evaluated, with the Cu-GA treatment achieving a maximum UPF of 128.43. Enhanced wash, rub, and light fastness in Fe and Cu mordanted samples were explained via coordination bonding between dye chromophores and fiber. These results demonstrate how machine learning and decision science tools can generalize analytical predictions in dye systems. The integrated ANN-WASPAS framework offers a transferable analytical strategy applicable to broader natural product formulations, quality control, and sustainable materials research.