A new SVM algorithm and selected application in sequence analysis

摘要

A new support vector machine (SVM) algorithm based on incomplete independent component analysis named decomposition forward SVM dimension reduction algorithm is presented. The computational complexity is lower than that of the traditional sub-block algorithm. We use the idea of incomplete ICA, with decomposition forward SVM, hence reducing the computational complexity. Using the proposed algorithm for two-class classification and multiclass classification of protein sequences, experimental results show that if reducing the dimension from 110-dimension to 5-dimension, the mean recognition rate is superior to the traditional sub-block algorithm. This shows the effectiveness of our proposed recognition system.