Neurowall-DNN: Gradient-Guided Defensive Neural Architecture for Multi-Class Network Attack Detection

The rapid proliferation of sophisticated multi-class cyberattacks has exposed critical limitations in conventional Intrusion Detection Systems (IDS), particularly in handling evolving attack patterns and zero-day threats. Most existing deep learning-based IDS models rely on static optimization strategies, leading to reduced adaptability and higher false alarm rates under dynamic network conditions. To bridge this gap, this paper proposes NeuroWall-DNN, a gradient-guided defensive neural architecture designed for adaptive and resilient multi-class network attack detection. The proposed framework integrates a deep neural network with gradient-based reinforcement feedback, enabling dynamic parameter adjustment and enhanced feature discrimination. Adaptive gradient optimization is employed to strengthen decision boundaries, while reinforcement-driven updates improve convergence stability and attack generalization. The model is evaluated using NSL-KDD and CICIDS2017 benchmark datasets under a multi-class classification setting. Experimental results demonstrate an overall accuracy of 98.4% on NSL-KDD and 99.1% on CICIDS2017, outperforming conventional DNN and hybrid IDS baselines in detection rate and false positive reduction. The proposed architecture achieves faster convergence and improved robustness against unseen attack categories, establishing NeuroWall-DNN as a scalable and intelligent defense mechanism for next-generation cybersecurity infrastructures.