Credit Card Fraud Detection based on CSat-Related AdaBoost
Yue Yang, Chenyuan Liu, Ningning Liu · 2019
In the field of Financial Technology, machine learning provides important support for decision-making through the effective use of data. Credit card fraud detection technology is a good example, but it still faces two challenges: the unbalanced data sets and cost-sensitive characteristics. In this paper, we proposed an enhanced CSat (Customer Satisfaction)-related AdaBoost. Based on the traditional AdaBoost, we consider the expected loss of the impact of customer satisfaction and re-adjust the weight of different categories in the cost adjustment function of the basic classifier. Considering the serious consequences of fraud transactions, we also implemented a metric related to the Total Profit of Classification (TPC) to evaluate performance. The results show that the CSat-related AdaBoost performed better in F1-score and AUC score compared to the traditional AdaBoost and some mainstream models, the reliability and interpretability of TPC as an evaluation metric is also demonstrated in our paper.