A Deep Convolutional Neural Network and SDN-based Closed-Loop System for Real-Time Network Attack Detection and Mitigation

The increasing sophistication of cyber threats demands more intelligent and adaptive network defenses than traditional architectures can provide. This research bridges this gap by designing and implementing an integrated system that leverages a Deep Convolutional Neural Network (DCNN) for real-time traffic classification within a Software-Defined Networking (SDN) control loop, enabling automated threat mitigation. Evaluated in an emulated environment, the system demonstrated a perfect 100% detection rate with zero false positives against a range of common attacks while introducing minimal operational overhead, with only a 3.1% throughput reduction. Crucially, it preserved legitimate throughput by a factor of 4.2x during attacks, proving the viability of a closed-loop, DCNN-SDN framework for achieving robust, self-defending network security without compromising performance.