Approximate Supplement-Based Neighborhood Rough Set Model in Incomplete Hybrid Information Systems

Incomplete hybrid information systems (IHISs) contain hybrid data (e.g., categorical data, numerical data) and incomplete data. With the development of big data, IHISs widely exist in various practical applications. Due to the heterogeneity of hybrid data and the complex semantics of incomplete data, effectively processing the IHIS has become a significant challenge. The established indiscernibility relations of the existing studies for dealing with IHIS over-amplify the uncertainty of missing values, which may achieve unsatisfactory results. In this paper, we propose an approximate supplement-based neighborhood rough set model (AS-NRSM) to deal with the data of IHISs. Specifically, we propose a method to approximate supplement missing values with known values or constructed interval values, and the original IHIS is becoming the constructed IHIS*. Then, we formulate a novel similarity function to construct the improved neighborhood tolerance relation and the corresponding neighborhood tolerance classes. Finally, we design two experiments on 5 UCI data sets by introducing three performance metrics. Experimental results illustrate that the proposed AS-NRSM has higher classification performance than the two representative models.