Comparison of k-nearest neighbor, quadratic discriminant and linear discriminant analysis in classification of electromyogram signals based on the wrist-motion directions

In this study, the authors compared the k-Nearest Neighbor (k-NN), Quadratic Discriminant Analysis (QDA), and Linear Discriminant Analysis (LDA) algorithms for the classification of wrist-motion directions such as up, down, right, left, and the rest state. The forearm EMG signals for those motions were collected using a two-channel electromyogram(EMG) system. Thirty normal volunteers participated in this study. Thirty features with a time-window size of 166 ms per feature during a 5-s forearm muscle motion were extracted from the gathered EMG signals. The difference absolute mean value (DAMV) was used to construct a feature map and the LDA, QDA, and k-NN algorithms were used to classify the directions of the signal. The recognition rates were 84.9% for k-NN, 82.4% for QDA, and 81.1% for LDA. There was a statistically significant difference between the k-NN and LDA algorithms (P < 0.05).

Research highlights
► Compared the k-Nearest Neighbor (k-NN), Quadratic Discriminant Analysis (QDA), and Linear Discriminant Analysis (LDA) algorithms for the classification of wrist-motion directions.
► Thirty normal volunteers participated in this study.
► The difference absolute mean value (DAMV) was used to construct a feature map.
► The recognition rates were 84.9%for k-NN, 82.4% for QDA, and 81.1% for LDA.
► We believe that k-NN is better than QDA and LDA as a classifier.