Electromyography (EMG) Signal based Knee Abnormality Prediction using XGBoost Machine Learning Algorithm

When diagnosing illness, assessing sports injuries, and rehabilitating patients, Surface Electromyography (sEMG) signals play a crucial role. In this study, we use the machine learning technique extreme gradient boosting (XGBoost) to detect knee abnormalities. Four primary sequential steps involved in predicting knee abnormality for a publicly available dataset with 22 subject’s sEMG signals were recorded from three different lower-limb muscles while they are walking, sitting, and standing activities where 11 appeared healthy and 11 had knee abnormalities based on clinical assessment. First, we apply wavelet denoising the sEMG signals to eliminate noise and apply segmentation with a window size of 256 ms, overlapped it by 25%. In the third stage, feature extraction is done using Time Domain(TD), later feature reduction method using principal component analysis (PCA). Finally, a machine learning algorithm is used to predict lower limb knee abnormalities. In this research, various Machine Learning models are implemented for prediction, like SVM, Naive Bayes, XGBoost, and KNN, where the XGBoost classifier gives a good accuracy of 96% compared to another classifier.