Formation of Fuzzy Classifier Structure by a Combination of the Class Extremum Algorithm and the Shuffled Frog Leaping Algorithm for Imbalanced Data with Two Classes
Marina Bardamova, I. A. Hodashinsky · Optoelectronics Instrumentation and Data Processing · 2021
Abstract A method of using the shuffled frog leaping algorithm as a tool for expanding the primary base of fuzzy classifier rules is proposed. This algorithm is relevant in the case when the existing rules are not sufficient for the qualitative recognition of all classes, for example, in the presence of data imbalance. Additional rules generated by metaheuristics can not only improve the quality of classification, but also provide a more complete description of the subject area under study. To create a compact initial structure of the classifier, the algorithm based on extreme values of features in classes was used. The studied combination was tested on 36 imbalanced data sets from the Knowledge Extraction based on Evolutionary Learning repository and showed an increase in the average geometric accuracy on 34 sets, as well as satisfactory results compared to similar algorithms. The advantages of the proposed method of forming the structure are the absence of necessity to augment the data with synthetic samples, low scatter of results on individual runs and the ability to improve the quality of classification by adding a small number of rules.