Principal Component Analysis Based on t-Distribution Hunting Search Algorithm

Abstract

Principal component analysis (PCA) is a linear dimensionality reduction algorithm in machine learning (ML), it plays an important role in feature extraction and noise removal. Its idea is to solve the maximum variance parameters of low dimensional spatial projection data by gradient descent (GD) method or stochastic gradient descent (SGD) method, so as to determine the optimal submanifold in low dimensional space. However, GD method or SGD method are easy to fall into the trap of local minimum, and are difficult to obtain the global optimal solution. In this paper, a t-distribution hunting search algorithm (THSA) with global optimization ability is proposed to replace the GD optimization method and improve the dimensionality reduction effect of PCA.