Many real-life data sets are incomplete, i.e., some attribute values are missing. Mining incomplete data sets is truly challenging. Among many methods of handling missing attribute values applied in data mining. We will discuss two approaches: rough sets combined with rule induction and the CART system based on surrogate splits. The main objective of this paper is to compare, through experiments, the quality of rough set approaches to missing attribute values with the well-known CART approach. In our experiments we used only lost value interpretation of missing attribute values.
展开▼
