판단자-조언자 방식의 머신러닝 모형을 통한 신용카드 부정 사용 탐지

초록

Various machine learning methods have been suggested for the detection of credit card fraud. Besides exploring single-algorithm models, hybrid models utilizing multiple algorithms together are also being proposed. Furthermore, data imbalance is a major issue in fraud detection, making it difficult to improve both false positive and false negative rates simultaneously. In this paper, a judge-advisor hybrid layered model is proposed with layers of advisor models and a judge model. The advisor layer includes multiple classification models composed of single algorithm for each model, while the judge layer utilizes its own algorithm and the prediction results of the advisor models as part of the input data. Comparing the performance of the overall model including the judge layer with various advisor layer models, values of precision and recall are not steady depending on the characteristics of the advisor model and the method of processing the results of advisor models. However it is confirmed that the judge-advisor model can improve precision and recall rates in the face of data imbalance through the appropriate selection of algorithms for the layer of advisor and judge and the method of processing prediction results of advisor models.

키워드