A SYSTEM FOR AUTOMATIC DETECTION AND PREDICTION OF FRAUD IN FINANCIAL TRANSACTION SEQUENCES

摘要

An automatic system for compiling and analyzing transaction sequences to classify them into similar groups is provided. The system also discovers frequently occurring sub sequences for use in fraud detection and prediction at an individual transaction level in financial industry. The system automatically compiles a full set of sequences from the transaction history of selected accounts. User can define a set of parameters for characterizing the transactions, and the system uses rules, business intelligence and learning networks to build characterizations for each of the found sequences. Fraud scores for each of the discovered sequences are computed, and all sequences are subjected to intelligent analysis using string matching, graph search and self organizing map techniques. All similar sequences are found, and clustered. Their conditional probabilities are calculated. All commonly occurring sub-sequences are also found, and their conditional probabilities are calculated. The derived set of probabilities and sub-sequences is used to modify, dynamically, fraud scores of all participating transaction sequences. Finally, predictive models are built using learning techniques and any new or standard pattern found using these predictive models are stored for quick access and reuse.