Performance of K-nearest neighbor algorithm and its variants on prediction of muscle fatigue and determining indicative features

The Anterior Cruciate Ligament (ACL) is a key ligament that contributes to knee stability. In recent years, there has been a significant increase in non-contact ACL injuries that result in knee instability. These injuries are often caused by fatigue in the knee. Thus, by detecting fatigue in athletes, coaches can adjust training regimens to prevent injuries and improve performance. In addition to that, in the medical field, fatigue detection can help doctors in identifying and treating conditions such as muscular dystrophy and multiple sclerosis. Fatigue detection from EMG signals can provide valuable insights and improve safety and performance in a variety of fields. In this study, various features that can be used to differentiate between fatigue and non-fatigue states were investigated. The study used electromyography (EMG) electrodes to record EMG signals to analyze muscle contraction during knee motion. Participants were asked to perform exercises until they reach maximum exertion. The resulting data were processed and 42 features were extracted. 11 features of importance were selected and then a comparative study for supervised machine learning algorithms as well as unsupervised machine learning methods was conducted. KNN gave the highest accuracy out of all the classifiers with the weight-adjusted KNN (WA-KNN) variant giving the highest accuracy of 93.7%. The contribution of the features in WA-KNN was analyzed and it was found that mobility, zero crossing, and maximal fractal length are the features that had the highest positive impact on the model. Those findings indicate that those three features are useful in predicting fatigue and preventing fatigue.