A Research On Different Feature Expression In OFFSS

Feature subset selection is a very useful and important problem for pattern recognition and machine learning. The aim is to delete the irrelevant or the redundant feature and select the optimal relevant feature in the feature space, thereby, improve the performance of the classification and recognition is very necessary. Optimal fuzzy-valued feature subset selection (OFFSS) method is proposed to deal with the problem of the expression of the imprecise data. OFFSS use the fuzzy set to express the imprecise data substitute the traditional discrete method. This paper provides a comparative research of replacing discrete value of a fuzzy value. It first introduces the OFFSS heuristic algorithm and then improves this algorithm to make it more perfect and with stronger capacity of the noisy tolerance. Second we analyze looking on the imprecise data as the fuzzy-valued feature will get more satisfied outcome than the discrete-valued feature base on the improved algorithm. Last we apply two methods in the fuzzy decision tree induction, the experimental outcome prove OFFSS method can get higher training accuracy and testing accuracy. The OFFSS system is implemented, the function of this system is consummate, it can well dispose the multi-class classified problem of database, and the ability of deal with the large-scale database is enhanced evidently.