Abstract:

Abstract To build an Intrusion Detection System (IDS) for identifying and categorising cyber-attacks in a prompt and autonomous manner both on the network and the host level, machine learning approaches are being utilized extensively. On the other hand, due to the fact that malicious attacks are always evolving and taking place in extremely high volumes, it is necessary to develop a solution that can be scaled up. The cyber security community has access to a variety of malware datasets that can be used for further research in the public domain. Furthermore, no study that is currently accessible has given a complete evaluation of the effectiveness of different machine learning techniques on different datasets that are publicly available. A form of hybrid deep learning model, Recurrent Neural Network-Long Short Term Memory (RNN-LSTM) is investigated in this paper with the goal of developing a flexible and effective intrusion detection system that can detect and classify unanticipated and unforeseen cyber-attacks using the datasets KDDCup99 and NSLKDD. The results of this type of study make it easier to select the optimal algorithm that has the potential to effectively work in identifying future cyber attacks. By using the KDDCup and NSLKDD datasets, a thorough analysis of experiments evaluating RNN-LSTM and other traditional machine learning classification models is shown. The abstract as well as high-dimensional feature representation of the intrusion detection system data is used to develop the RNN-LSTM model that was developed by feeding the features into a large number of hidden layers. Two experiments using binary and 5- group classification method has been performed, it is observed from these experiments that the performance of RNN-LSTM method based on binary classification is 83.29% and 68.59% using KDD Cup 99 and NSLKDD dataset which is better as compared to other conventional machine learning classifier and 5-group classification method.

Keywords: