A Comparison of classification accuracy for gender using neural networks and discriminant funtion analysis based on biomechanical measures of isokinetic torque, work, power, fatigue index, countermovement jump and acceleration

Neural networks can be applied to many predictive data mining applications due to their power, flexibility and relatively easy operations. Predictive neural networks are very useful for applications where the underlying process is complex, such as in classification using a mix of nominal and ratio level variables and for predictive validity based on classification modelling. A neural network can approximate a wide range of statistical models without requiring the researcher to hypothesize in advance certain relationships between the dependent and independent variables. Neural networks and discriminant function analysis (a more traditional statistical approach) were applied, based on a set of dependent variables, which were biomechanical measures of isokinetic peak torque, work, power and fatigue index for leg extension at 60 °s^-1, 180 °s^-1 and 300°s^-1; a counter movement jump or CMJ (contact time, flight time and peak height); and I Om acceleration were collected within a university exercise physiology laboratory, and were compared for their classification accuracy to successfully discriminate between physically active men and women. In this research gender represented the classification variable in the statistical analyses. Neural networks, specifically the multilayer perceptron (MLP) networks, were applied and compared with stepwise method discriminant function analysis for classification accuracy. Sixty participants (male=32, age=23.9, s.d.=8.0 years; female=28, age=21.1, s.d.=6-3 years) represented the volunteer/convenient sample in the study. The torque, work, power and fatigue index data were normalised by dividing these scores by the participants weight to minimise weight-muscle mass differences that are related to gender. Normalisation refers to the division of multiple sets of data by a common variable in order to negate or diminish the variable's effect on the data, in this context weight, thus allowing underlying characteristics of the data sets to be compared and results in the derivation of force/weight, work/weight and power/weight ratios. The research aim focussed on which factors or variables provided the greatest difference between the genders, the establishment of a hierarchy of factor-variable importance and to assess which multivariate method of classification provided the best solution. The stepwise discriminant method was applied to derive a statistical meaningful solution in terms of variable hierarchy of importance. The neural network classification statistics indicated the order of importance was peak torque 180 °s^-1, contact time, flight time and peak torque at 300 °s^-1 and the accuracy of classification was I 00% on the training model but only 85% for the holdout/testing sample. Of the input variables in the stepwise discriminant analysis four were identified as statistically significant and were in the following order of importance, peak torque 180°s^-1, CMJ contact time, fatigue index at 180 °s^-1 and total work 180°s^-1 and the percentage classification accuracy was 90.6% males and 96.4% females ( overall 93.3%). Neural MLP networks were more effective in predicting group membership and displayed higher predictive validity when compared to discriminant analysis. The selected variables in both models were very similar with peak torque 180 °s^-1 identified as the most important discriminating factor/variable for gender differences for normalised test scores for torque, work, power, and CMJ and acceleration motor fitness tasks.