Predictive models and parameter analysis for multiple tactile perceptions in skin-wet fabrics interface

The tactile perception of wet fabrics plays an important role in human comfort, but the underlying mechanisms of skin-wet fabric interaction remain unclear. This study measured the basic properties (saturated water content, wicking height, moisture regain rate, fabric density, and yarn diameter) of 20 fabrics, collected real-time tactile physical parameters (skin cooling rate, normal pressure, frictional coefficient, maximum acceleration amplitude, and acceleration mean square deviation), and subjective tactile perception ratings (wetness, coldness, roughness, stiffness, and total hand feeling value) during the interactions between fingertip and fabrics with three relative water contents (dry, 35%, and 70% saturated water content). The results demonstrated the significant effects of fabric water content on tactile perceptions and tactile physical parameters. Four machine learning methods, Gradient Boosting Regression, XGBoost Regression, Random Forest Regression, Artificial Neural Networks Regression (ANNR), and a linear regression method, Partial Least Squares Regression, were used to establish predictive models. The ANNR model demonstrated the best performance in total hand feeling value (R² = 0.727, root mean square error = 0.66), and SHapley Additive exPlanations analysis revealed eight key factors, highlighting the significant effect of fabric water content and mechanical stimulation on total hand feeling value. These findings provide deeper insights into enhancing the comfort of wet textiles and offer potential guidance to optimize the tactile design of products.