A Novel Technique to Defraud Credit Card by Handling Class Imbalance Problem Using Machine Learning

Abstract

Typically, classification algorithms work poorly when confronted with unbalanced datasets, and the resulting effects are skewed against the majority class. As a result, an effective model is required to identify unbalanced data, particularly in the context of fraud detection. For these types of issues, the classifier's accuracy is not trusted because the cost of predicting a fraud sample as a non-fraud sample is extremely high. In general, imbalanced learning happens when some types of data distributions significantly outnumber other data distributions in the instance space. There is a need of technique such as under sampling or oversampling in order to learn from unbalanced datasets. A novel over sampling method has been suggested for learning from unbalanced datasets in this paper. The basic impression here is that a weighted distribution for diverse outnumbered class instances has been utilized depending on their degree of complexity to learn, with more pretended evidence leading to the outnumbered ones, being more troublesome to learn. As a result, with regard to data distributions, the suggested approach improves learning first by bringing down the bias familiarized using class difference, and then by pliantly conveying the classification judgement boundary toward challenging instances.