Comparative analysis of sEMG signal classification using different K-NN algorithms

Abstract

Bio-medical signal based techniques are having highly successful rate for designing the robotic
rehabilitation system due to its robustness for controlling the robotic device. In this context, the
classification accuracy plays a pivotal role. This paper presents the results of elbow movement as
well as four fingers movement classification by employing the K-Nearest Neighbors (K-NN)
algorithms with Discrete Wavelet Transform (DWT) and Principle Component Analysis (PCA). In this
scenario, DWT is applied for de-noising as well as feature extraction purpose whereas PCA is used
for the data size reduction. In DWT, Daubechies 4 (db4) wavelet filter is deputed at fourth level and
fourth level approximation coefficient of reconstructed signal (a4) is utilized for time-frequency
feature extraction. The result exhibits that Fine K-NN has 95.67% classification accuracy which is
found to be best among the other K-NN algorithms.