Data’s hidden data: Qualitative revelations of sports efficiency analysis brought by neural network performance metrics

The study explores the technical optimization of an athlete through the use of intelligent system performance metrics that produce information obtained from inertial sensors associated to the coach's technical qualifications in real time, using Mixed Methods and Machine Learning. The purpose of this study is to illustrate, from the confusion matrices, the different performance metrics that provide information of high pertinence for the sports training in context. 2000 technical fencing actions with two levels of complexity were performed, captured through a single sensor applied in the armed hand and, simultaneously, the gesture’s qualification through a dichotomous way by the coach. The signals were divided into segments through Dynamic Time Warping, with the resulting extracted characteristics and qualitative assessments being fed to a Neural Network to learn the patterns inherent to a good or poor execution. The performance analysis of the resulting models returned a prediction accuracy of 76.6% and 72.7% for each exercise, but other metrics indicate the existence of high bias in the data. The study demonstrates the potential of intelligent algorithms to uncover trends not captured by other statistical methods.