No abstract

Electromyography (EMG) signal is one of the important tools for the detection of skeletal related muscular information. In this scenario, we analyse the normality and abnormality of knee movements for pre-predicting the knee dystonia diseases. It is characterized by involuntary sustained muscle contractions affecting one or more sides of the body, frequently causing twisting and repetitive movements or abnormal postures. In this paper, EMG based knee dystonia pre-prediction is assessed by employing EMG-Lower Limb dataset. Then, transformation is done by using discrete wavelet Transformation (DWT), which delivers a superior time and frequency localization ability. After transformation, hybrid feature extraction is performed by employing yule-walker, burg's, voltage level (minimum and maximum), Renyi entropy, and Peak-Magnitude to Root Mean Square Ratio (PMRS) for achieving optimal feature subsets and also to reject the redundant and irrelevant features. In most of the existing studies, an individual feature or two different features are combined for extracting the feature values from the acquired signal. In this research, five effective entropy and autoregressive features are combined to obtain more active features. After obtaining the active features, a superior binary classifier: Kernel Nearest Neighbour (KNN) is implemented for classifying the normality and abnormality of knee movements. The experimental outcome proves that the proposed methodology effectively distinguishes the normal and abnormal knee movement in terms of sensitivity, specificity, recall, precision, accuracy and E-rate. The proposed methodology improves the classification accuracy in knee movement detection upto 0.785-1.65% compared to the existing methods.

No abstract