Improving PLS-RFE based gene selection for microarray data classification

摘要: Gene selection plays a crucial role in constructing efficient classifiers for microarray data classification, since is characterized by high dimensionality and small sample sizes contains irrelevant redundant genes. In practical use, partial least squares-based gene approaches can obtain subsets of good qualities, but are considerably time-consuming. this paper, we propose to integrate squares based recursive feature elimination (PLS-RFE) with two schemes: simulated annealing square root, respectively, speed up the process. Inspired from strategy schedule, proposed eliminate number features rather than one informative during each iteration removed decreases as proceeds. To verify effectiveness efficiency approaches, perform extensive experiments on six publicly available three typical classifiers, including Naive Bayes, K-Nearest-Neighbor Support Vector Machine, compare our ReliefF, PLS PLS-RFE selectors terms classification accuracy running time. Experimental results demonstrate that accelerate process impressively without degrading more compact both two-category multi-category problems. Further experimental comparisons subset consistency show approach scheme not only has better time performance, also obtains slightly root scheme. We classify data.Two dynamic schemes combined PLS-RFE.The select similar PLS-RFE.Experimental their actual use.