Smart classification of normal and aggressive muscle actions, Normal ve Agresif Kas Hareketlerinin Akilli Siniflandirmasi

Abstract

In the present study, linear, non-linear and statistical approaches so named Fourier Correlation (FC), Wavelet Correlation (WC) and Pearson Correlation (PC), respectively have been compared to each other in estimating cross-correlations between two EMG series, simultaneously collected from the same muscle groups (biceps, triceps, thighs) on symmetric limb (legs and arms). The features, which are computed for both normal and aggressive muscle actions, are computed for four electrode pairs and then these features are classified by different classifiers which are Penalized Logistic Regression (PLR), Random Forests (RF), Conditional Inference Tree (CIT) and Support Vector Machines (SVM) with 10-fold cross-validation in order to have highest calculation accuracy (CA). Experimental data including six normal and six aggressive actions of 4 young participants (3 male and 1 female with the mean age of 21.8) is provided by the data base of UCI (University of California Irvine). PC has given the highest calculation accuracy (100%) with R programming for the RF classifier. CA obtained with this classifier and correlation method can be directly suggested to detect diagnostic evaluation and neuromuscular diseases and dysfunction. © 2018 Elsevier B.V., All rights reserved.