Evolutionary Instance Selection Algorithm based on Takagi-Sugeno Fuzzy Model
Sang-Hong Lee, Joon Shik Lim · Applied Mathematics & Information Sciences · 2014
Abstract: In this study, we propose evolutionary instance selection based on the Takagi-Sugeno (T-S) fuzzy model. The previous neural network with weighted fuzzy membership functions (NEWFM) supports feature selection; thus, it enables the selection of minimum features with the highest performance. The enhanced NEWFM supports a weighted mean defuzzification in the T-S fuzzy model with a confidence interval in the normal distribution; thus, it enables the selection of minimum instances with the highest performance. The enhanced NEWFM has two stages; feature selection is performed in the first stage, whereas instance selection is performed in the second stage. The performance of the enhanced NEWFM is compared with that of the previous NEWFM. In addition, McNemar’s test reveals a significant difference between the performances of both NEWFMs (p<0.05).