An Improved Intrusion Detection System based on convolutional neural networks (CNN) algorithm

Abstract

Abstract The evolution of the Internet has given rise to novel forms of network attacks, underscoring the importance of investigating methods for detecting anomalous behaviour and accurately classifying attacks within the domain of cyberspace. Recently, the Intrusion Detection System (IDS) has been employing various deep learning-based methodologies to construct a model that is driven by data. The aforementioned techniques are advantageous as they reduce the cost and labour involved in manual detection. The precision and capability of feature extraction in contemporary intrusion detection techniques exhibit significant variability. Furthermore, the presence of extraneous and repetitive phrases is ubiquitous in real-time network data. To tackle the aforementioned concerns, the present study proposes the utilisation of the convolutional neural networks (CNN) algorithm for the purpose of detecting intrusions. Initially, in order to enhance resource utilisation and reduce time complexity, a configuration of 16 layers and 8 convolutional neural networks with Max-Pooling1D is utilised. Subsequently, a series of evaluative experiments conducted on the NSL-KDD dataset indicate that the enhanced CNN model has the potential to enhance the performance of the Intrusion Detection System (IDS). The precision for all types of features is reported to be 0.991, while the recall and f1-measure are noted to be 0.995 and 0.993, respectively