An AI-Powered Web App for Winter Footwear Slip Risk Estimation-Pilot Study

Slip and fall incidents represent a major public safety and economic burden. While standardized mechanical testing methods such as ASTM F2913 provide laboratory-based coefficient of friction measurements, these approaches are often less reliable than human-centered assessments. This gap poses critical challenges for footwear manufacturers aiming to optimize traction designs without relying on costly and time-consuming human-centered testing. Therefore, developing predictive models to estimate human-centered slip resistance score represents a crucial advancement in footwear safety assessment. This pilot study presents a novel approach that combines regression modeling with deep learning-based feature extraction to predict human-centered slip resistance scores on both dry and wet ice surfaces. The methodology integrates mechanical test outcomes with visual outsole and material characteristics to reduce prediction error compared to individual models. Nested cross-validation was used to optimize hyperparameters. The fused model achieved superior performance with RMSE of 1.20° ± 0.29° and MAE of 0.78° ± 0.28° on dry ice, and an RMSE of 1.40° ± 0.45° with an MAE of 1.02° ± 0.31° under wet ice conditions, substantially outperforming models trained on a single data modality. To translate this advancement into practice, the final model is deployed as a web application.