A Fuzzy-Rough-Based Approach for Uncertainty Classification on Hybrid Information System

Hybrid information system (HIS) contains variety types of data including boolean, categorical, missing, real-valued, and set-based data, which are becoming very important for meaningful information analysis in real applications. The defect of the previous classification algorithms mainly lies in that the feature types of dataset haven't been comprehensively considered. In this paper, the internal relations among objects based on features that can reflect multiple data types have been expressed by heterogeneous distance function. And then, a new fuzzy-rough relation calculation under HIS and a new fuzzy-rough lower and upper approximation have been analyzed. Finally, the fuzzy-rough nearest neighbor classification under HIS (FRNN-HIS) algorithm has been proposed. Eight datasets has been adopted to conduct experiments. The experiments have shown that the proposed FRNN-HIS algorithm can significantly outperform FRNN and VQNN algorithms in terms of predicting classification results.