A novel extreme learning machine for dimensionality reduction on finger movement classification using sEMG

Abstract

Projecting a high dimensional feature into a lowdimensional

feature without compromising the feature characteristic is a challenging task. This paper proposes a novel dimensionality reduction constituted from the integration of extreme learning machine (ELM) and spectral regression (SR). The ELM in the proposed method is built on the structure of the unsupervised ELM. The hidden layer weights are determined randomly while the output weight is calculated using the spectral regression. The flexibility of the SR that can take labels into consideration leads a new supervised dimensionality reduction called SRELM. Generally speaking, SRELM is an unsupervised system in term of ELM yet it is a supervised system in term of dimensionality reduction. In this paper, SRELM is implemented in the finger movement classification based on electromyography signals from two channels. The experimental results show that the SRELM can enhance the performance of its predecessor, spectral regression linear discriminant analysis (SRDA) because it has better class separability than SRDA. In addition, its performance is better than principal component analysis (PCA) and comparable to uncorrelated linear discriminant analysis (ULDA).