A Novel Hybrid Approach Based on Rough Set for Classification: An Empirical Comparative Study

Uncertainty learning is an essential research direction of rough set theory. Uncertainty is defined as a situation with inadequate information and includes three types: inexactness, unreliability and ignorance; it is not merely the absence of knowledge. However, uncertainty can still exist in cases where there is a big chunk of information. In this regard, Rough Set Theory (RST) is a powerful mathematical model that deals with uncertain information. The RST offers a practical method for extracting decision rules from datasets. To deal with the uncertainty within the datasets, we initially propose a new hybrid approach that combines RST with Machine Learning (ML) algorithms. Furthermore, a novel algorithm based on RST (RSTPNN) is proposed by the idea of combining probability based Naive Bayes (NB) and nearest neighbor based K-Nearest Neighbors (KNN) to efficiently improve the classification performance. For experimental validation, we use 12 different datasets from the Open ML and UCI repositories. The empirical results evidently show that the proposed method may improve classification performance.