One-pass-throw-away learning algorithm based on hybridization of LDA and PCA

This paper proposes a new learning algorithm based on the versatile elliptic basis function (VEBF) by considering only the most data distributions for automatic computing the appropriate width vector. In addition, the orthonormal basis and Linear Discriminant Analysis (LDA) technique are also applied to the proposed method for adjusting the directions of the hyperellipsoid in the network and improving the performance. After tested by real world data sets, the proposed method illustrates that it outperforms the VEBF and other learning algorithms. © 2013 IEEE.