Construction Of The Detection Model Of Abnormal Transactions Of Bank Customers In The Abnormal Detection Equipment

Along with our country ditching cash payment industry,bank card application scenario is gradually expanding,which made our country bank transaction scale in the era of big data,at the same time the abnormal trading volume has been growing rapidly,not only to the banking industry has brought great risks,and caused the heavy losses directly or indirectly.Therefore,the construction of the detection model of abnormal transactions of bank customers in the abnormal detection equipment is the key to solve these problems.The recognition model of bank account abnormal transaction based on rules and general machine learning is always not satisfactory when learning more complex serialized transaction features,and it needs to spend a lot of resources to summarize transaction business and update rules on a regular basis,resulting in resource consumption and increased delay.In this paper,based on the study of the domain knowledge of banking risk control and the analysis and research of machine learning algorithm,is proposed to trade trinity of anomaly detection,from three different angles respectively analyze the deal,the first perspective focuses on trading behavior analysis of the same bank card user groups,according to the homogeneity is the common feature of group to accurately judge the trading of individual behavior deviation as much as possible.The second aspect mainly focuses on the comparison with the user's own historical trading behavior,and on the comparison with the customer's historical trading behavior in different trading periods,so as to accurately discover the characteristics and rules of the user's historical trading behavior and identify obviously different trading behaviors.The third Angle starts from the correlation between different trading subjects to find out the internal relationship between direct or indirect trading objects,and discover the implicit abnormal relationship between trading subjects.And need to go to extract respectively from the Angle of different analysis is used to identify the abnormal trading behavior characteristics,according to different Angle detection need user transaction behavior characteristics,respectively using the cluster analysis,neural network,and other technical data path discovery algorithm training study,combined with the bank of the user's basic information,according to the three sub models respectively their test results,the final vote mechanism fully integrated all aspects of anomaly information and give the final discriminant of abnormal test results and completes the model building.Through the above methods,the disadvantages such as weak learning ability of serialized transaction features and limited learning ability of features in each transaction record can be effectively avoided,so as to improve the operation efficiency and performance of the bank account abnormal transaction recognition model.