Open AccessRunning and Cycling Induced Fatigue on Wrapper vs. BLR Feature Selection for IBk Classification
Author(s) -
Siling Tang,
Wei Ping Loh,
Chien-Chuan Lin
Publication year - 2019
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/530/1/012060
Subject(s) - cycling , accelerometer , feature selection , overfitting , computer science , artificial intelligence , archaeology , artificial neural network , history , operating system
Running and cycling fatigue causes muscle pains, cramps and accidental injuries. Previous studies had considered the importance of tri-axial accelerometer to detect fatigue motion in stability, balance and postural deviation aspects. While tri-axial accelerometer is important, the capability to predict running and cycling fatigue from the biomechanical attributes were unclear. Therefore, the study aims to (i) compare the featured attributes selected from wrapper approach and Binary Logistic Regression (BLR) on running and cycling datasets and (ii) perform IBk classification accuracy comparison on the feature selection attributes. Public running, experimental running and cycling induced fatigue datasets were employed to test the analysis. The most significant attributes identified in the public running was RMS_ML, followed by Range_ML and the cycle frequency in experimental running and cycling respectively. On 10 folds cross-validation classification test using the IBk algorithm in WEKA, accuracies for experimental running and cycling datasets were 93.1% and 90.5% from wrapper method, 65.6% and 76.2% from BLR respectively. Wrapper method performs better than BLR in data overfitting phenomenon. Findings reveal that the mediolateral variation at body trunk motion plays a major impact to predict fatigue running but fatigue cycling shows cycling frequency as the main attribute in fatigue cycling prediction.