Rule Extraction from Incomplete Decision Tables

Rule extraction is an important issue of data mining and many efficient algorithms based on rough sets have been presented for obtaining rules from decision tables. However, little work has been focused on extracting rules from the incomplete decision tables. In this paper based on an improved discernibility matrix an efficient method for obtaining all optimal credible decision rules from an incomplete decision table is proposed. Through uniting the objects of a maximal tolerance class into a new object the scale of discernibility matrix used to produce the disjunction of rules is greatly reduced, and then the computation efficiency of the rule extraction gets an obvious improvement. Theoretical analysis and experiments indicate that the improved method is more efficient for obtaining optimal credible decision rules from an incomplete decision tables.