Hybrid Chicken Swarm Optimization (CSO) and Fuzzy Logic (FL) Model for Handling Imbalanced Datasets

Abstract

The unequal representation of the various divisions existing in the data is one of the main data inconsistency issues. Data with an imbalanced distribution negatively influence the efficiency of most conventional classifiers. This paper introduces a new method for over-sampling the handling of imbalanced data sets. The method hybridize chicken swarm optimization and fuzzy logic (CSO-FL). The proposed model ensures that the synthetic samples generated reside in minority regions. The proposed hybrid CSO-FL applied on three datasets with different imbalanced ratios between 1.78 and 129.44. It demonstrated significant improvements in the efficiency of various classification techniques. During the classification process, we used KNN, DT, SVM and Naïve classifiers. The obtained results were very promising; the precision, sensitivity, and F_score values are enhanced in all classifiers. The values in one dataset improved with ratios >90 % in many classifiers because of the high imbalanced ratio in this dataset, while in the other two datasets, the measurement values enhanced with ratios from nearly 10% to 30%. The CSO-FL approach compared with three different approaches on the same datasets. The approaches are SMOTE algorithm, modified SMOTE, and TGT algorithm and proved to outperform their results.