Deep Q-ICAN: A deep reinforcement learning-based approach for real-time CPA attack detection and mitigation in NDN architecture

Abstract

Named Data Networking (NDN) has emerged as a transformative architecture for next-generation Internet design, leveraging its data-centric paradigm to address scalability and security challenges inherent in traditional host-centric networks. By prioritizing content naming, stateful forwarding, and in-network caching, NDN inherently enhances efficient data dissemination and trust through cryptographic content validation. However, its reliance on distributed caching introduces vulnerabilities to Cache Pollution Attacks (CPA), where malicious actors inject illegitimate content to disrupt caching efficiency, degrade data availability, and inflate retrieval latency. Proactive mitigation of CPA is critical to preserving NDN's performance benefits and ensuring reliable content delivery, as unchecked attacks undermine consumer trust and network resilience. In this paper, we present Deep Q-ICAN: a deep reinforcement learning based intrusion detection and prevention system for NDN. It is an online, and adaptive approach designed to effectively mitigate CPA attacks and enhance the caching mechanism within the NDN architecture. Through extensive experiments in diverse and realistic topologies, we show that our solution achieves superior performance compared to previous approaches in the same field. In fact, Deep Q-ICAN achieves an accuracy of 98.87% and an Average Cache Hit Ratio of 80%. It improves the caching strategy by approximately 42% while maintaining the states of the NDN routers resources. Furthermore, Deep Q-ICAN significantly improves the Average Retrieval Delay (ARD) from 0.284s to 0.065s, thereby enhancing the efficiency of retrieving desired content from the Content Store (CS) for legitimate consumers.

Keywords Cache Pollution Attack, Deep reinforcement learning, Future internet, Intrusion detection, NDN