User Session Modeling for Effective Application Intrusion Detection

With the number of data breaches on a rise, effective and efficient detection of anomalous activities in applications which manages data is critical. In this paper, we introduce a novel approach to improve attack detection at application layer by modeling user sessions as a sequence of events instead of analyzing every single event in isolation.We also argue that combining application access logs and the corresponding data access logs to generate unified logs eliminates the need to analyze them separately thereby resulting in an efficient and accurate system. We evaluate various methods such as conditional random fields, support vector machines, decision trees and naive Bayes, and experimental results show that our approach based on conditional random fields is feasible and can detect attacks at an early stage even when they are disguised within normal events.