Modelling traction of studded footwear on sports surfaces using neural networks

Traditional regression techniques have shown limited used in the development of empirical models for the traction performance quantities of studded footwear on surfaces. This is due to the unknown and often non-linear relationship between performance parameters, such as traction force, and input variables, from the shoe and surface. Experimental data has been used to train artificial neural networks to model the relationship between stud parameters, namely cross-sectional area, length and two shape coefficients, with dynamic traction as the output variable. A variety of neural network structures and optimisation algorithms were evaluated. The most promising network gave an average prediction error of 10% compared to an error of 36% when an optimised linear model is employed. This study shows that the neural network technique as powerful potential in understanding the effect of shoe and surface parameters and in the optimisation of traction forces experienced by athletes.