Multivariate Analysis of Determining Factors for Athlete Performances in Judo

Abstract

The present study aims at classifying judo athletic levels using multivariate analysis of physical and technical data. A sample of 42 judo athletes from two competitive groups (28 national level and 14 state level) was submitted to the following tests and measurements: (a) skinfold thickness; (b) circumferences; (c) breadths; (d) stabilometric test; (e) Special Judo Fitness test; (f) dynamometry test. Logistic regression (LR) and multilayer perceptron neural network (MLP) were employed to determine variables that best classify the two groups. The classifiers select seven variables and both LR and MLP models presented similar performances with 90.0% and 91.0% accuracy, respectively. These results suggest that a reduced set of biomechanical, anthropometric and physiological variables allow to assess the athletic level of judo players.