Intrusion detection system: a deep neural network-based concatenated approach

In recent years, the field of information security has seen a substantial rise in the use of approaches that include deep learning. The implementation of deep learning strategies into intrusion detection systems has proven to be rather successful. In this study, we use an optimization strategy to provide a concatenated learning model that is based on three different convolution neural network (CNN) models. These models are VGG16, VGG19, and Xception. The experiment was carried out using two different benchmark datasets, namely UNSW-NB15 and CIC DDoS 2019. During pretraining, images were converted into square color images suitable for CNN model applications using a feature selection technique based on the information gain value. The findings of the experiment indicate that the accuracy that can be achieved by combining an optimal deep learning strategy with the concatenation method proves to be superior. The result obtained from the proposed intrusion detection model has an optimum accuracy of 99.26% and 96.23% in both the CIC DDoS 2019 and UNSW-NB15 dataset, respectively.