A Preliminary Study on ReliefF based Feature Ranking for Classification of Myoelectric Signals

Myoelectric pattern recognition is an important technique in the design of prosthetic devices. The two important parameters in the working of the myoelectric pattern recognition technique are feature extraction and feature ranking. The computation of excess feature extraction results in redundancy. In this study; 12 time domain features were extracted. To optimize the number of features for the classification of phases of pick and place task, a ReliefF feature ranking algorithm is used. The subjectwise classification was performed using the k-NN classifier. From the classification outcomes; the evaluation of the features was performed using 5 levels effective chart. It was found that Auto Regressive coefficient and Willison Amplitude were the most effective features, followed by the Waveform Length, Root Mean Square and Myopulse Percentage Rate. The tabulation of features in the designed chart will ease the processing time and procedure in the design of the myoelectric pattern recognition process.