Evaluating the Efficacy of Deep Learning Models in Network Intrusion Detection

Abstract

As for the field of network security, the creation of an effective and reliable Network Intrusion Detection System (NIDS) task is still a challenging issue. Traditional techniques are not very effective as they are mostly based on signature-based identification and do not work well with new, complex threats. The following are limitations that this research seeks to overcome by developing an enhanced hybrid deep learning ensemble model that combines different machine learning (ML) and deep learning (DL) algorithms. Logistic Regression, Stochastic Gradient Descent (SGD), LightGBM, XGBoost, and a Deep Neural Network (DNN) are integrated in a hybrid model to improve the detection performance by using the stacked ensemble technique. It is proved that the comprehensive cooperation of these models outperforms the results of any single model with the accuracy of 0.982. This superior performance is evidenced through performance evaluation such as radar and line graphs.

The findings show that the present ensemble method enhances IDSs’ effectiveness and reliability, providing a holistic strategy for addressing the issues arising from the constant evolution of current networks. As for the future work, the enhancement of the ensemble model will be continued, the detection in real-time will be investigated, and the application of the proposed methodology to other areas of cybersecurity will be investigated.