Application of unsupervised machine learning for performance-driven segmentation of sustainable knitted fabrics of ultraviolet protection and moisture-related comfort

This study presents a data-driven methodology for the comprehensive assessment and segmentation of the multidimensional comfort performance of sustainable knitted fabrics. We investigated the combined effects of fiber composition (Hemp and Refibra™-based yarns) and diverse knit geometries (plain, purl/links-links, and moss stitch) on five key functional parameters: ultraviolet protection factor, air permeability, water vapor permeability, drying rate, and wicking height. To accurately classify the fabrics based on holistic performance, the technique for order preference by similarity to ideal solution was first employed as a multi-criteria decision-making tool to establish a reliable three-tiered reference classification consisting of high, medium, and low comfort levels. This dataset was subsequently used to validate the performance of various unsupervised machine learning algorithms, including k-means, Gaussian mixture models, spectral clustering, and hierarchical clustering. The hierarchical clustering algorithm demonstrated superior efficacy and stability, achieving the highest agreement rate of 0.72 against the established multi-criteria reference. The resulting optimal segmentation successfully isolated the high-comfort cluster, identifying the T1-2L and T4-2L structures as top performers. These samples were characterized by an exceptional balance between ultraviolet protection and moisture/air transport capabilities. The overall findings confirm that while the material composition significantly affects liquid transport, the links-links structure is critical for achieving multi-functional performance.