Relational Frequent Patterns Mining for Novelty Detection from Data Streams

We face the problem of novelty detection from stream data, that is, the identification of new or unknown situations in an ordered sequence of objects which arrive on-line, at consecutive time points. We extend previous solutions by considering the case of objects modeled by multiple database relations. Frequent relational patterns are efficiently extracted at each time point, and a time window is used to filter out novelty patterns. An application of the proposed algorithm to the problem of detecting anomalies in network traffic is described and quantitative and qualitative results obtained by analyzing real stream of data collected from the firewall logs are reported.

• Type: Conference
• Year: 2009
• Where: MLDM
• Authors: Michelangelo Ceci, Annalisa Appice, Corrado Loglisci, Costantina Caruso, Fabio Fumarola, Carmine Valente, Donato Malerba
• Keywords: Consecutive Time Points, Frequent Relational Patterns, Machine Learning, MLDM 2009, Time Point