Detecting Anomaly in the Usage of Database Attribute.

In database applications, database operations should be provided to maintain persistency of the data which is often represented as database attributes. Any missing, redundant or inconsistent operation performed on database attributes would indicate anomaly or even program bugs. Through characterizing operations performed in database transactions on database attributes, we extract a feature vector from code for each attribute. This paper proposes a clustering-based approach which analyzes the feature vectors to automatically detect anomalies in the usage of database attributes. Once an anomaly is detected, developers can perform investigation to take corrective actions if necessary. The evaluations on both industrial and open source database applications show that our approach is able to detect many types of anomalies in the usage of database attributes with a high detection rate (92.8% on average), and a low false positive rate (0.57% on average). Keywords—anomaly detection; database application; clustering; attribute usage.