A Hybrid Method of Feature Subset Selection

抄録

Feature subset selection is of prime important in pattern classification, machine learning and data mining applications. A real world database may contain many noisy, unnecessary and irrelevant features. If it is used for data minihg directly, the quality of the discovered knowledge may be very poor. To cope with this problem, many methods have been proposed. In this paper, we propose a hybrid algorithm by using class mutual information for feature selection, starting from the Rough Sets CORE. If the CORE is empty we use binary mutual information for the first feature selection. Experiments have been conducted on some artificial and real world domains in terms of tree size, test errors rate and subset sizes. The results show the effectiveness of proposed hybrid algorithm.