배영랑  이준헌  김태경

63권  1호, pp. 18-29, 2월  2026
10.12772/TSE.2026.63.018

superhydrophobic dyes Polypropylene Molecular descriptors Dyeability Machine Learning XAI

PDF Full-Text

Polypropylene fibers are difficult to dye because of their extreme hydrophobicity and high crystallinity. Therefore, we have developed and reported superhydrophobic dyes which have high affinity towards polypropylene fibers. In the previous studies, dyeability (K/S) was analyzed using a single molecular descriptor (logP) which effectively represented the hydrophobicity of the dyes. However, as more and more dyes were synthesized and their dyeing results obtained, the dyeability could no longer be explained by the single molecular descriptor alone. In order to numerically analyze the relationship between the dye structures and dyeability on polypropylene fibers, a machine learning approach based on multiple molecular descriptors was applied. Linear regression and random forest models were used, and model interpretation was performed using weight analysis and SHAP for explainable artificial intelligence (XAI). Both models achieved high redictive performance, indicating that the relationships between molecular descriptors and dyeability were successfully captured. Model interpretation revealed that logP, degree of alkyl substitution, and hydrogen-bond donors were consistently identified as the key molecular descriptors affecting dyeability. These results demonstrate that dyeability is determined by the combined effects of multiple molecular characteristics, providing a quantitative basis for the rational design of dyes for polypropylene fibers.