ATM交易状态特征分析与异常检测

Analysis of ATM Transaction State Characteristics and Abnormal Detection

针对ATM交易系统产生的交易数据,对交易量、交易成功率和响应时间进行特征分析进而建立异常检测模型。利用方差分析法以及3σ思想,均值设为u,标准差设为σ,将每天交易量与均值的差值在u±3σ之外的值标记为异常点。对于成功率和响应时间,首先利用k-means聚类分析将数据进行划分,然后利用决策树思想确定类之间的阈值,根据阈值可区分出数据异常点。由此可找出交易量、交易成功率和响应时间的异常值,进而建立ATM异常检测系统。

Based on the transaction data generated by ATM transaction system, the characteristics of trading volume, transaction success rate and response time are analyzed and the anomaly detection model is established. By using the variance analysis method and 3σ ideas, set for the mean u, standard deviation σ, mark the difference between the daily trading volume and the mean value over u±3σ as an abnormal point.In terms of success rate and response time, firstly, k-means cluster analysis is used to divide the data and then use the decision tree to determine the threshold value between classes, and the abnormal data points can be distinguished according to the threshold value.According to the above method, the abnormal value of trading volume, transaction success rate and response time can be found, and the ATM anomaly detection system can be established.