Network Intrusion Detection with Deep Neural Networks for Enhanced Cybersecurity

Abstract

The increasing sophistication of cyberattacks has challenged the effectiveness of traditional signature-based intrusion detection systems, which rely heavily on predefined attack patterns. This study aims to develop and evaluate a Deep Neural Network (DNN)-based approach for network intrusion detection to enhance cybersecurity performance. The proposed model was trained and tested using two benchmark datasets NSL-KDD and CICIDS2017 following a systematic data preprocessing process, including normalization, feature encoding, and data partitioning. The DNN architecture employed multiple hidden layers with ReLU activation and Adam optimization to capture complex, non-linear traffic patterns. Experimental results demonstrated that the DNN model achieved accuracy levels of 98.6% on NSL-KDD and 99.2% on CICIDS2017, with corresponding high precision, recall, and F1-scores. The confusion matrix and ROC curve analysis further confirmed the model’s capability to accurately distinguish between normal and attack traffic, with an AUC value of 0.995, indicating superior classification performance. Comparative evaluation showed that the DNN significantly outperformed traditional signature-based systems by reducing false positives and effectively identifying novel attacks. In conclusion, the findings highlight the DNN’s potential as a robust and adaptive framework for modern network intrusion detection, capable of improving detection accuracy, operational efficiency, and resilience against evolving cyber threats.