Research On Screening For Abnormal Bank Transaction Based On Scan Statistics

The stability of the financial environment is one of the major driving forces in the development of the modern economy.However,there are various of illegal trade in the world,such as money laundering and illegal fund-raising,that are endangering the stability of the world financial system.At the same time,bank transactions are one of the important ways for illegal activities,regulation and testing of bank transactions have drawn increasing public attention.Therefore screening of abnormal transaction information from huge bank transaction data efficiently and accurately has become the urgent demand of the current banks and financial regulators.Facing the problem,a new abnormal transaction detection method: scan statistical analysis method is introduced in this paper,and the traditional scan statistics method is improved to apply to bank transaction data.The specific research content is as follows.First of all,based on the shortcomings of the existing research and the characteristics of bank transaction data,a statistics of transaction data—the amount of transaction information is constructed in this paper,which including account ID,transaction type,transaction date and transaction amount that are selected from the153 fields of bank transaction data.After data processing,a specific formula for the amount of transaction information is constructed in this paper,as the main body of follow-up research.To avoid mistakes due to a single study of the transaction amount of bank transaction data.Secondly,scan statistics is introduced to filter out the trading range with abnormal aggregation.Combining with the characteristics of the specific banking transaction data and the lack of traditional scan statistics methods,the scan statistics methods is improved,using the Monte Carlo simulation and log-likelihood ratio test to replace the calculation of the distribution or approximate distribution of scan statistics,which avoids the difficulty of calculating the distribution of scan statistics.When the transaction data presented a unimodal form,a scan statistic method based on Poisson distribution was proposed.Then,the applicability of the scanning statistical analysis method to the screening of abnormal bank transactions and the superiority of the improved scan statistics method are verified.Finally,for the bank transaction data with bimodal form,a scan statistical model based on the double Poisson distribution is established.At the same time,Monte Carlo simulation and log-likelihood ratio test in the improved scan statistic model are also introduced.Then using the real transaction data to conduct an empirical analysis of bank abnormal transaction screening,it is verified that the double Poisson distribution scanning statistical model has better applicability and higher accuracy in the abnormal bank transaction screening problem when the bank data presents a bimodal form.