Chronohunt: Determining Optimal Pace for Automated Alert Analysis in Threat Hunting Using Reinforcement Learning

Threat hunting stands out as a proactive practice applied to identify stealthy threats that evade traditional detection mechanisms. Although powerful, threat hunting demands significant investments in terms of knowledge, time, and resources to meticulously analyze massive amounts of logs, and formulate threat hypotheses. Particularly, real-time threat hunting necessitates substantial manpower and computational resources to identify threats and might lead to inefficiencies and overlooked threats. Conversely, while more economical in resource allocation, batch-mode hunting risks missing fast-moving threats. To address these pivotal challenges, we formulate the problem of pacing the threat hunting in security operational environments and design Chronohunt, a solution that automatically and adaptively adjusts the threat hunting pace in alignment with the security importance, volume of events, available resources, and the evolving threat landscape. Chronohunt integrates two optimizations: (i) an initial heuristic optimization using grid search to establish a baseline hunting pace, and (ii) a dynamic optimization using reinforcement learning to dynamically fine-tune the pace in response to changes in the environment (e.g., hunting performance, evolving threat landscape, event importance, etc.). Obtained results show the efficacy of Chronohunt in adaptively aligning the hunting pace based on changes in the environmental conditions while ensuring high accuracy in threat hunting and optimal resource utilization.