Improving SMOTE Technology for Credit Card Fraud Detection Category Imbalance Issues

Credit cards play an important role in today's economy, but they also provide fraud conditions for outlaws. Often, the data for fraud detection is extremely imbalanced, which seriously affects the detection effect of classification models. The KCSMOTE (Kmeans Center Synthetic Minority Oversampling Technique) model is proposed to address the problem of imbalance in credit card fraud data affecting the effectiveness of model detection. The K-means algorithm is used to cluster the samples to find safe clusters with different sparsity, and then K-means++ is used to find a few class centroids of the safe clusters, using the centroid as the base points to improve the SMOTE algorithm. XGBoost and Random Forest algorithms were used to validate the effectiveness of the KCSMOTE model. ADASYN, k-means-SMOTE, borderlineSMOTE, SMOTETomek, SMOTEEnn, and SMOTEWB as well as the original data were selected for comparison experiments, and several metrics, F1-score, Precision, Recall, and AUC (area under the curve), were chosen to determine the results. Experimental results show that the KCSMOTE model is more effective in dealing with unbalanced fraud data than other sampling algorithms.