ATM交易状态特征分析与异常检测

Signature Analysis and Anomaly Detection of ATM Transaction Status

通过累计图判断ATM交易状态各变量的周期性,计算以工作状态、时段为周期的季节指数,通过直方图和非参数检验判断各变量的分布规律。随后,将交易变量Z标准化,依据主成分分析法提取该组变量的主元,计算每个交易时刻的主元得分,根据常规控制图和离群值判断系统运行异常或故障时刻,判处出主元得分处于异常的时刻,以应用状态为因变量,建立含工作状态、交易时段两个虚拟变量的多元逻辑回归方程。最后,对模型的解释能力、模型显著性、回归系数显著性进行检验。

First of all,the periodicity of variables of ATM transaction status was judged through cumulative graph.The seasonal index which takes the working condition and time period as cycle was calculated.The distribution of variables was determined through the histogram and non-parametric test.Next,the trading variable Z,was standardized and the main element of this group of variables were extracted according to the principal component analysis.The score of the main element at each trading moment was calculated.Then the abnormal moment which the score of the main element implies was determined according to the operating abnormal andfault moment of conventional control charts and outlier judgment system.Thirdly,taking the application status as dependent variable,a multiple logistic regression equation was established with two dummy variables,including working status and trading time.Finally,The explanatory power of the model,the significance of the model and the significance of the regression coefficient were tested.