Improving network security through deep learning RNN approach

Subject of the Study. This article explores the use of Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, to improve the effectiveness of Intrusion Detection Systems (IDS). The study emphasizes the work of optimizers to enhance the accuracy of detecting network attacks and provides a comparative analysis of various optimization algorithms using the NSL-KDD dataset. Method. The article proposes an RNN-LSTM-based approach for detecting intrusions in network traffic. Seven different optimization algorithms were evaluated, including Adamax, SGD, Adagrad, Adam, RMSprop, Nadam, and Adadelta. The method involves a comparative analysis of their performance with varying hidden layer sizes. Main Results. The experiment methodology included training the RNN-LSTM model with hidden layer sizes ranging from 50 to 100 over 500 epochs. The Adamax optimizer achieved the highest accuracy of 99.79%, while Adadelta had the lowest accuracy at 97.29%. Additionally, SGD demonstrated the best True Positive Rate (TPR), while Adamax showed the lowest False Alarm Rate (FAR). The study evaluated metrics such as accuracy, TPR, FAR, precision, and F1-score, with Adamax standing out for its overall performance. Practical Significance. The article is relevant to professionals in the fields of cybersecurity, network security, and IDS development. This article provides valuable insights for enhancing IDS configurations to improve the detection and mitigation of network intrusions.