Imbalanced Data Classification Using Combination of Oversampling and Fuzzy Support Vector Machines

ABSTRACT Classifying imbalanced data stands as a critical aspect in machine learning, posing substantial hurdles due to the uneven distribution of data. Diverse methods have emerged to address such challenges in data categorization. This study aims to alleviate data imbalances while leveraging Fuzzy Support Vector Machines (FSVM) to bolster resilience against noisy and outlier data in mining tasks. Initially, our approach involves preprocessing the data via the SMOTE algorithm to establish a balanced dataset. This algorithm synthesizes data for the minority class by considering the proximity of individual samples. Following this, we employ Fuzzy Support Vector Machines to classify the preprocessed data. Lastly, we introduce a novel membership function for FSVM. The UCI dataset serves as the testing ground. Comparative results showcase the proposed method s adeptness in effectively handling imbalanced data.