FADS: Flocking anomalies in data streams

Anomalies detection in data has gained a lot of attention in many domains due to the increasing number of attempts of fraud. In this paper, a new multi-agent approach to detect anomalies in data exploiting a clustering algorithm, is proposed. Each data item is associated with an agent and the agents are randomly disseminated onto a virtual space where they move following the flocking algorithm. The agents grouping in flocks based on a well-defined concept of similarity of their associated objects. The agents associated with similar objects grouping in flocks, whereas the agents associated with objects dissimilar to each other do not group in flocks. The objects associated with agents do not grouped in flocks represent the anomalies in data. Features of the proposed approach, such as parallelism, asynchronism, and decentralization, makes the algorithm scalable to very large data sets. Experimental results confirm the validity of the FADS algorithm for real and synthetic datasets.