The present disclosure discloses the explain fraud detection method and system via fraud graph and features similarity based kNN few-shot example-prompts (EPs). In the present disclosure, a transaction is initiated by the transaction cards and the transaction data is received by a feature engineering module (203), which extracts meaningful information from the transaction data. The data extracted by the feature engineering module (203) is then received by the deployed module (300) that identifies whether the transaction is a fraudulent transaction, and according to the transaction fraud, example prompts (EPs) are generated. Said generated example prompts (EPs) are few-shot example prompts (EPs) and based on the kNN technique applied on the graph similarity and the vectors (features) similarity, the top k example prompts (EPs) are segregated. The explanation module receives the prompt data and segregates and characterizes the prompts based on the importance score. The explanation module is a large language module (LLM) which generates a human-readable explanation to tell clients why this transaction is labelled as a fraudulent transaction.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.