Detecting APT attacks using an attack intent-driven and sequence-based learning approach

Advanced persistent threats (APTs) are a significant threat to network security as they can disintegrate the security fortress of enterprises. Recent studies have focused on detecting APT attacks by matching typical tactics, techniques, and procedures (TTPs) associated with APT attacks. However, the lack of positive APT samples affects the performance of existing approaches. To address this challenge, we propose a novel attack intent-driven and sequence-based learning approach (AISL) for APT detection. AISL integrates heterogeneous audit data and creates corresponding security tags based on attack intent. Specifically, we investigate various data sources of attack detection and establish a dedicated network event ontology. Based on this ontology, we construct a provenance graph that integrates audit data from heterogeneous sources. During the construction of the provenance graph, we identify and tag potential attack behaviors based on attack intent to increase the number of positive samples in the dataset. Finally, we train a tag-sequence-based semantic model for APT detection. We evaluated AISL through ten realistic APT attacks and achieved an average precision of 93.05%, recall of 98.12%, and F1-score of 95.36%, outperforming state-of-the-art approaches.