A Weighted Fuzzy Rough Nearest Neighbor Classification Algorithm Based on Multiple Interpolation and Similarity Attribute Analysis

Abstract

Upper and lower approximation of fuzzy-rough set membership degree is used to solve uncertainty of classification problem in FRNN (Fuzzy Rough Nearest Neighbor) algorithm. Although FRNN is the current leading classification algorithm, misjudgments still tend to occur when handling similar attribute values. Combining multiple interpolation algorithms and similarity attribute analysis, this paper proposes a new classification algorithm, which is called weighted Fuzzy Rough Nearest Neighbor (WFRNN) classification algorithm. WFRNN adds the corresponding weight of each attribute for the sample, and then multiple interpolations are used to fill data sets and the other four kinds of packing method are adopted to fill the missing data set. Then five completely random missing data sets from UCI were used in comparison experiments. We have compared WFRNN with classic KNN, decision tree, FRNN, J48, and random forests. Experimental performances show that the WFRNN algorithm can predict more accuracy classification results.