Credit card fraud detection is exposed to problems such as large-scale sample data, high computational complexity, and extremely unbalanced data distribution. To solve those problems, this study proposes a convolutional neural network (CNN) and utilizes large-scale credit card transaction data to detect fraud. At the same time, considering the extremely unbalanced transaction data, the K-means algorithm is employed for clustering and is combined with support vector machine synthesis minority oversampling technology (SVMSMOTE) to increase the number of minority samples. Finally, a KM-SVMSMOTE-CNN-based prediction model for credit card transaction fraud is built, and the credit card fraud data released on the Kaggle platform is selected for verification. The experimental results show that the fusion model based on KM-SVMSMOTE-CNN greatly improves the overall recognition rate of credit card fraud detection.