Classification Predictive Models of Running- and Cycling-Induced Fatigue

The lower limb muscle fatigue and changes in lower limb kinematics during running and cycling indicate early signs of fall and injury risk. Past studies had analyzed the effects of running and cycling fatigue using the EMG or accelerometers. However, no study has developed the fatigue classification predictive models in running and cycling motions. The objectives of this study were to (i) distinguish pre-fatigue and fatigue condition in running and cycling motions and (ii) evaluate the accuracy of fatigue classification predictive models. Two cases studies involving 40 participants (14 males 6 females, (22.5 +/- 3.9 years old); 16 males 4 females (26.1 +/- 4.3 years old)) performing incremental treadmill run and cycle ergometer tests were carried out to induce fatigue. Measurements consisted of accelerometry, sacral trajectory and lower limb kinematics. The significant attributes contributing to binary classes (pre-fatigue/fatigue) were identified using the wrapper and information gain approaches. The binary logistic regression and decision tree classification predictive models were developed. The models were made up of 4 to 5 significant attributes, providing the strong evidence that the fatigue model classifications were more accurate than the IBk Regression and J48 classifications. Findings revealed 56.8-81.1% classification accuracies for the binary logistic regression and decision tree models. However, the binary logistic regression model gives 24.3% and 12.5% higher classification accuracies as compared to the decision tree model for running and cycling respectively. The regression model also outperforms the IBk Regression by 8.1% in running and 2.5% in cycling dataset. This study gives an indication that the binary logistic regression is a superior predictor of running and cycling fatigue classification.