Classification performance of motor unit action potential features

Abstract

The objective of this study is to examine the classification performance of the following motor unit action potential (MUAP) feature sets: i) time domain measures, ii) frequency measures, iii) autoregressive coefficients AR, and iv) cepstral coefficients. Two different feature selection methods were used: i) univariate analysis, and ii) multiple covariance analysis. Both methods showed that: i) the duration measure is the best discriminator, ii) the median frequency, FMED is the best discriminator among the frequency measures, and iii) the cepstral coefficients are better discriminators than the AR coefficients. Furthermore, the recognition rate of the above feature sets was investigated using the K-means nearest neighbour clustering algorithm. Time domain measures and cepstral coefficients gave the highest recognition score.