Rough set and CART approaches to mining incomplete data

摘要

Many data sets are incomplete, i.e., are affected by missing attribute values. In this paper, we report results of experiments on two approaches to missing attribute values. The first one is based on rough set theory and rule induction, the second one is the CART method that uses surrogate splits for handling missing attribute values and that generates decision trees. As follows from our experiments, both approaches are comparable in terms of an error rate. Thus, for a specific data set the best method of handling missing attribute values should be selected individually.