With the rapid development of Internet technology, the convenience of the financial system has been greatly improved, but criminals have also used network technology to implement more diversified financial fraud, of which credit card fraud has become a key problem to be solved in the banking industry. Traditional anti fraud systems typically rely on a combination of expert experience and machine learning algorithms to identify fraudulent behavior by summarizing its characteristics. However, in the face of increasingly complex, diverse, and gang like fraudulent methods, this approach gradually exposes its shortcomings. Therefore, it is imperative to establish a more precise and efficient credit card anti fraud system. This article introduces the knowledge graph theory and proposes a new anti fraud analysis model to address the problem of imbalanced black sample datasets in credit card fraud, which makes it difficult to effectively analyze individual relationships and high-risk group characteristics. The knowledge graph is used to mine the target entities associated with black samples, such as ID numbers, devices, IP addresses, and phone numbers. By extending the association, a network containing black samples is constructed, and high-risk groups are identified through community segmentation to extract their network and individual relationship features. On this basis, a credit card anti fraud system based on knowledge graph was constructed, which utilizes the full amount of credit card application data and third-party data for group division, feature analysis, and risk ranking, outputs high-risk group information, and uses individual relationship features as inputs to the machine learning model to further improve the prediction accuracy of the model.