Improving Network Attack Detection Using Hybrid Feature Selection and Deep Learning Methods

Early detection of network attacks is the primary security challenge in cloud applications, smart cities, and the industry's sixth revolution.Signature-based or anomaly-detection models detect network attacks.Innovative attacks pose significant challenges to signature-based detection methods.These methods rely on predefined attack patterns, making them ineffective against novel threats, though they remain widely used due to their efficient resource utilization.In contrast, anomaly-detection models can identify such attacks by analyzing deviations from normal network behavior but often require substantial computational resources, limiting their practicality in resourceconstrained environments.To address this gap, this research proposes a hybrid algorithm combining Feature Selection (FS) methods with Deep Neural Networks (DNNs) to develop an efficient anomaly-detection model.FS reduces data dimensionality, optimizing resource usage, while DNNs enhance detection accuracy for innovative attacks.The resulting hybrid model aims to balance the low resource consumption of signature-based methods with the advanced detection capabilities of anomaly-based systems, offering a practical solution for real-world applications where both performance and efficiency are critical.The paper applied eleven FS methods to the NSL-KDD dataset to investigate the best features for attack detection.Subsequently, two DNNs were developed using a grid search hyperparameter tuning method to determine the most compatible model for the selected features.The Hybrid DNN and PERMUTATION FEATURE importance achieved optimal detection with minimal complexity.The results show a gain in detection accuracy by 2% to 24% and computational cost by 47.5% to 60%.