DIFFERENTIATING sEMG SIGNALS UNDER MUSCLE FATIGUE AND NON-FATIGUE CONDITIONS USING LOGISTIC REGRESSION CLASSIFIERS

In this work, an attempt has been made to differentiate surface electromyography signals under fatigue and non-fatigue conditions. Signals are recorded from the biceps brachii muscles of 50 healthy volunteers. A well-established experimental protocol is followed for this purpose. Signals are subjected to further processing and features namely amplitude of first burst, myopulse percentage rate, Willison amplitude, power spectrum ratio and variance of central frequency are extracted. Three types of logistic regression classifiers, linear logistic, polykernel logistic regression and multinomial regression with ridge estimator are used for automated analysis. Classifier parameters are tuned to enhance the accuracy and performance indices of algorithms, and are compared. The results show distinct values for extracted features in fatigue conditions which are statistically significant (0.0027 = P = 0.03). All classifiers are found to be effective in demarcating the signals. The linear logistic regression algorithm provides 79% accuracy with 40 iterations. However, in the case of multinomial regression with ridge estimator, only 7 iterations are required to achieve 80% accuracy. The polykernel logistic regression algorithm (0.06 = ? = 0.1) also provides 80% accuracy but with a marginal increment (1 % to 4 %) for precision, recall and specificity compared to other two classifiers.