An attribute reduction approach and its accelerated version for hybrid data
Wei Wei, Jiye Liang, Yuhua Qian, Feng Wang · 2009
In practical issues, categorical data and numerical data usually coexist, and a unified data reduction technique for hybrid data is desirable. In this paper, an information measure is proposed for computing the discernibility power of a categorical or numeric attribute. Based on the measure, a uniform definition of significance of attributes with categorical values and numerical values is proposed. Furthermore, an algorithm to obtain an attribute reduct from hybrid data is presented, and one of its accelerated version is also constructed. Experiments show that these two algorithms can get the same reducts, and the classification accuracies of reduced datasets are similar with the ones using Hu's algorithm. However, the accelerated version consumes much less time than the original one and Hu's algorithm do.