Integrating footwear features into fatigue prediction models for marathon runners: A hybrid CNN-LSTM approach

Abstract

Footwear design, especially the curvature of carbon plates, may influence fatigue perception, but few studies have integrated footwear features into fatigue prediction models. This study aimed to develop a hybrid CNN-LSTM model to predict runners’ fatigue states and evaluate the impact of footwear characteristics on fatigue perception. Twelve male marathon runners (age = 21.8 ± 1.3 years; body mass = 59.1 ± 4.1 kg; height = 168.9 ± 2.2 cm; and weekly mileage = 68.8 ± 5.5 km) participated. They wore two types of carbon-plated shoes (flat plate, FP, and curved plate (CP)) and ran at a steady pace (Borg score 13) until a Borg score of 16 or 85% of maximum heart rate was reached for 2 min. EMG signals and physiological data were collected during treadmill running. A hybrid CNN-LSTM model was trained with and without footwear features to predict fatigue states. The model with footwear features achieved 85% accuracy, compared to 69% without. Curved carbon plate (CP) shoes delayed semi-fatigue onset, indicating better initial support, but the time to full fatigue was similar for both shoe types. The CNN-LSTM model effectively predicted fatigue states, with significant improvement when footwear features were included. Footwear design, particularly carbon plate curvature, influenced fatigue perception.