Comparative Analysis of Machine Learning Techniques for the Classification of Knee Abnormality

Knee abnormality is a major problem in elderly people these days. It can be diagnosed by using Magnetic Resonance Imaging (MRI) or X-Ray imaging techniques. X-Ray is only used for primary evaluation, while MRI is an efficient way to diagnose knee abnormality, but it is very expensive. In this work, Surface EMG (sEMG) signals acquired from healthy and knee abnormal individuals during three different lower limb movements: Gait, Standing and Sitting, were used for classification. Hence, first Discrete Wavelet Transform (DWT) was used for denoising the input signals; then, eleven different time-domain features were extracted by using a 256 msec windowing with 25% of overlapping. After that, the features were normalized between 0 (zero) to 1 (one) and then selected by using the backward elimination method based on the p-value test. Five different machine learning classifiers: K-nearest neighbor, support vector machine, decision tree, random forest and extra tree, were studied for the classification step. Our result shows that the Extra Tree Classifier with ten cross-validations gave the highest accuracy (91%) in detecting knee abnormality from the sEMG signals under analysis.