Anomaly Detection in Transactional Sequential Data

A transactional system often has a steady distribution based on their decisions or responses on various transactions. If the system is changed, the distribution also often changes. It is a valuable work to detect anomaly caused by system changes based on different distributions. In this study, we modeled those decision or response signals into a series of time-related distributions and then proposed a method combining distance metrics and anomaly detection to discover whether changes have happened in some systems. Distance metrics on different distributions can decide whether changes have happened and anomaly detection can find what happened further. Extensive experiments show that our method has a good performance and can locate the anomaly accurately.