Identifying Significant Attributes Features for Treadmill Running Fatigue Classification

抄録

The effects of treadmill running impact on the runner’s fatigue state were often examined from interactions among corresponding attributes. While fatigue accompanied by tired legs, muscle pain and cramps were often described, no study has classified the levels of treadmill running motion into fatigue and non-fatigue conditions using data mining approach. This study identifies the significant attributes that promote accurate classification of treadmill running fatigue conditions. The study data was retrieved from a benchmark public database that exhibits treadmill running variability. The qualitative and quantitative features of anthropometric, demographic, accelerometry and sacral trajectory attributes were initially evaluated for data classification. Five classifiers: Lazy, Function, Meta, Rules and Tree built-in Waikato Environment for Knowledge Analysis (WEKA) tool were employed on the training and 10 folds cross validation test modes. The [CorrelationAttributeEval] and [WrapperSubsetEval] were applied for attribute evaluation in order to enhance classification accuracies at 10 folds cross validation mode. This study distinguishes significant attributing data features into fatigue and non-fatigue levels on average classification accuracies for the before and after significant attribute considerations: Lazy (28.3%, 63.3%), Function (56.4%, 59.3%), Meta (45.9%, 60.0%), Rules (46.5%, 59.0%) and Trees (48.2%, 65.0%). Findings also reveal that the RMS of mediolateral is the most significant attribute whose variations exert a major effect on the treadmill running fatigue classes.