Modeling and prediction of cotton/polyester ring-spun yarn hairiness using Taguchi analysis, MLR, ANN and ANFIS

Abstract Yarn hairiness affects downstream processing and fabric appearance. This study evaluated five machine variables, front-roller cot hardness, front-to-middle roller gauge, front-roller pressure, apron spacer (clip) size, and spinning angle, for 30/70 cotton/polyester ring-spun yarn using a Taguchi L27 design. S3 hairiness was measured in five repetitions per run and analyzed by mean response, smaller-the-better signal-to-noise ratio, and ANOVA. Cot hardness was dominant, followed by front-roller pressure and spacer size; roller gauge had a smaller significant effect, whereas spinning angle had the smallest effect (parametric p = 0.050; permutation p = 0.019). The preferred combination produced the lowest L27 response (18.3 hairs/m), while an independent confirmation production gave 18.5 ± 0.85 hairs/m ( n = 5; CV = 4.59%). MLR, ANN, and ANFIS were compared using a common 24-run training and three-run test split. ANFIS gave the lowest held-out RMSE (0.273 hairs/m), whereas ANN gave the highest test correlation. Repeated six-fold cross-validation over ten repetitions provided a more stable assessment across data partitions, with ANN showing the lowest mean RMSE (0.554 ± 0.194 hairs/m), compared with MLR (0.834 ± 0.064 hairs/m) and ANFIS (0.845 ± 0.543 hairs/m). Given the small test set, model differences are interpreted descriptively.