The paper explores polynomial kernel function K(X, Y) = (X ?Y +1)?in support vector machine (SVM) by using information-geometrical method. In order to enlarge volume element of polynomial kernel function and then improve support vector machine classifier, we modify polynomial kernel function by applying a quasi-conformal transformation to it. The paper gives the expression of volume element of polynomial kernel function and mathematical proof for the case P = 1. The paper makes a study of introns and exons in DNA. By using window-sliding method, we propose a concept called relative difference ratio a(1(r)). It reflects the frequency difference of window sequence 1(r) appearing in intron sequence and exon sequence. According to a(l(r)), we obtain 47 feature sequences of intron and exon. A feature expression of DNA sequence is given based on these feature sequences. In the experimental part, we first get feature expression of sequences and then apply support vector machine with polynomial kernel function to the classification between intron sequences and exon sequences and the classification between exon sequences and genes. The results are good. The contrasting experiment between polynomial kernel function and modified polynomial kernel function shows remarkable improvement of the performance of support vector machine classifier, supporting our method.
|