Discretization algorithm based on difference-similitude set theory

A new discretization algorithm based on difference-similitude set theory (DSST) is presented. It is different from the known algorithms because the reduction in the information system is prior to the data discretization. Our algorithm discretizes the attribute values by analyzing difference sets of the newly-built decision table. The discretization is separated into three steps. The first step is to reduce the information system without any discretization. The second step is to rebuild a new decision table according to the reduction result obtained in the first step. The third step is to discretize the attribute values by analyzing the difference sets of the newly-built decision table. We applied our algorithm in UCI databases and compare our method with other algorithms at the end of the paper. The experiments show that it is an effective algorithm.