Hybrid fuzzy genetics-based machine learning with entropy-based inhomogeneous interval discretization
Yūji Takahashi, Yusuke Nojima, Hisao Ishibuchi · 2014
Discretization of continuous attributes is a key issue in classifier design from numerical data. In the machine learning community, continuous attributes are discretized into intervals. An entropy measure is often used to determine the cutting points for interval discretization. In the fuzzy system community, continuous attributes are usually discretized into overlapping fuzzy sets. Learning and optimization techniques are used to adjust the membership function of each fuzzy set. One interesting research issue is a comparison between interval partitions and fuzzy partitions. We address this issue by using an entropy-based interval discretization method in hybrid fuzzy genetics-based machine learning (GBML). Our hybrid fuzzy GBML algorithm is applied to a number of data sets where interval discretization is fuzzified with different fuzzification grades from zero (i.e., interval partitions) to one (i.e., completely fuzzified partitions). Experimental results from various fuzzification grades are compared with each other.