Handling data incompleteness using Rough Sets on multiple decision systems

A practical problem that arises in data analysis is to handle missing attribute values in an information system that has suffered degradation, so as to retain its quality. In this paper, we present a new Rough Set (RS) based approach to deal with incomplete data. The core idea is to tap the redundant information garnered from different databases that share common attributes. The attribute suffering missing entries in a deficient database is recast as a decision attribute in another reference database. The tenets of RS theory are then applied to derive rules that predict the missing values. Experimental results on pairs of two different pairs of related databases taken from the UCI repository reveal that our approach could predict missing values with a high degree of accuracy giving an average error of 15.75%.