Neighborhood components analysis in sEMG signal dimensionality reduction for gait phase pattern recognition

Dimensionality reduction technique is an essential method for sEMG signal pattern recognition and classification, especially for real-time application such as prosthesis control. This technique can reduce the high dimension extracted feature into a lower dimension space feature which help the classifier works more properly. This paper presents an application of a dimensionality reduction technique called neighborhood components analysis (NCA). We evaluate the efficiency of NCA by comparing its class separability and the classification accuracy with other three algorithms: principle component analysis (PCA), linear discriminant analysis (LDA) and local preserving projection (LPP). The result shows that NCA outperform other algorithm in the class separability, and its classification accuracy is also slightly higher.