Anomaly detection through quasi-functional dependency analysis

Anomaly detection problems have been investigated in several research areas such as database, machine learning, knowledge discovery, and logic programming, with the main goal of identifying objects of a given population whose behavior is anomalous with respect to a set of commonly accepted rules that are part of the knowledge base. In this paper we focus our attention on the analysis of anomaly detection in databases. We propose a method, based on data mining algorithms, which allows one to infer the "normal behavior" of objects, by extracting frequent "rules" from a given dataset. These rules are described in the form of quasi-functional dependencies and mined from the dataset by using association rules. Our approach allows us to consequently analyze anomalies with respect to the previously inferred dependencies: given a quasi-functional dependency, it is possible to discover the related anomalies by querying either the original database or the association rules previously stored. By further investigating the nature of such anomalies, we can either derive the presence of erroneous data or highlight novel information which represents significant exceptions of frequent rules. Our method is independent of the considered database and directly infers rules from the data. The applicability of the proposed approach is validated through a set of experiments on XML databases, whose results are here reported.