Document Type



Athletes’ movement biomechanics are of high interest to predict injury risk, especially in maximum effort cutting manoeuvres. However, using a standard optical measurement set-up with cameras and force plates influences the athlete’s performance. Therefore, alternative methods, e.g. Neural Networks, have been used to predict kinetic parameters based on easier to measure kinematic parameters. A previous study has evoked the question, whether the filtering processes of the input and output parameters used for training a feedforward neural network affect the prediction accuracy. To answer this question, four different filter combinations have been used during the pre-processing of joint angles, ground reaction force and joint moments of fast cutting manoeuvres, which were used to train a feedforward neural network. The results revealed a dependency.