Recurrent Neural Network Based Hybrid Deep Learning Architecture for Enhanced Network Intrusion Detection

Growing cybercrime in an era of widespread internet access presents major challenges to network security, especially in the realm of intrusion detection. While deep learning and machine learning have revolutionized network intrusion detection by helping identify between attacks and regular network traffic, it is still difficult to detect a wide variety of attacks with high accuracy. Additionally, this task becomes more challenging due to the computational demands of analyzing largescale data. To address these challenges, we have proposed a novel approach utilizing a Long Short-Term Memory (LSTM) based recurrent neural network, integrated with Convolutional Neural Network (CNN) architecture. The effectiveness of this model is assessed in our research in comparison to other deep learning techniques, such as basic RNN, LSTM, and GRU. Through extensive experimentation, our proposed hybrid CNN-LSTM model demonstrates superior performance, achieving 98.34% accuracy in multi-class classification and 98.7% accuracy in binary classification, with the most optimized publicly accessible CSE-CIC-IDS2018 dataset. Promising outcomes are also seen in important categorization metrics including F1-score, Precision, and Recall. Our results highlight the effectiveness of the integrated hybrid CNN-LSTM infrastructure as a strong classifier for cyberattack identification.