Research On Feature Selection Algorithm In Pattern Classification

The information we obtained explodes with the rapid development of technology.How to get valuable information from these data becomes one of the hot spots.Artificial intelligence such as pattern classification emerged gradually. In order toobtain maximum information related to certain characteristics, pattern classificationsystem will get the maximum information through the following four steps. First, getcharacteristic data of something. Second, preprocess to the data followed by featureselection, and finally get the characteristics evaluation of characteristic data. Correctclassification of some specific things will be got after these four steps. However, theemergence of more and more high-dimensional data will result in the curse ofdimensionality, irrelevant and redundant features. The traditional algorithms need tobe optimized and innovated so that new algorithm is more versatile stronger andmore efficient in the run. The important part of feature selection is a prerequisite forus to get good classifier for pattern classification system.This paper describes the theoretical basis and feature selection classification offeature selection algorithm. Then the paper summarizes the domestic and foreignresearch status in feature selection algorithm. The author optimizes and innovates thealgorithm on the basis of ReliefF algorithm and PCA algorithm then creates theSecond feature selection algorithm based on optimized ReliefF algorithm and KPCAalgorithm. Firstly more effective information can be obtained from ReliefF-PCAalgorithm compared to applying ReliefF algorithm. Then the result of lowerdimension can be verified using KPCA algorithm than using PCA algorithm. Finally,the second feature selection algorithm and method of removing redundant featureswill be added in the algorithm so that it can effectively deal with the dimensionswhich are too high, with redundant features and characteristics unrelated data. Theexperiments show that the algorithm has strong classification accuracy.