Next-Gen Cyber Defense: Integrating Deep Learning into Threat Detection Systems

Abstract

Network traffic is evolving faster and cyber threats are increasingly becoming more sophisticated thus requiring more adaptive and efficient intrusion detection systems (IDSTraditional rule-based or signature-based intrusion detection system designs produce a significant number of false positives because they are unable to identify novel or sophisticated attack methods. The integration of deep learning methods such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Feedforward Neural Networks (FNN) to enhance network threat detection is examined in this paper. The study relies on CICIDS 2017 dataset, which is a rich dataset and includes a broad spectrum of attack types, such as Denial of service (DoS), Distributed Denial of service (DDoS), and Port Scanning. We evaluate the work of single models and a hybrid one of FNN, CNN and RNN to enhance the feature extraction and sequence analysis. Accuracy, precision, recall, F1-score, and AUC-ROC are the key measures of evaluation, used to evaluate the performance of the models. These results show that CNN model is the optimal one because it has accuracy of 98.66%, precision of 95.47%, recall of 96.63%, and AUC-ROC of 0.9990. FNN also performs well with an accuracy of 98.64 and RNN lower convergence although achieves an accuracy of 95.84. The hybrid model integrates the capabilities of FNNs, CNNs and RNNs and offers the similar outcomes. These findings demonstrate the potential of deep learning to detect both current and new cyber threats and the CNN model and FNN were observed to produce the overall best outcomes. In further research, it is possible to concentrate on more optimization (adversarial training and transfer learning) to obtain more specifics regarding detection and address the challenges of sequential processing of data.