An Optimized Deep Learning Framework for Network Intrusion Detection System (NIDS)

Abstract

Securing the network and associated infrastructure is a never-ending process. Researchers and network administrators are continuously developing new tools and technologies to protect the network and related infrastructure. Intrusion Detection System (IDS) is one of them. Artificial neural networks can learn and improve its performance if appropriately trained with the optimal feature vectors. In this paper, a Recurrent Neural Networks (RNN) and LSTM based machine learning model is proposed to detect and classify the type of intrusion depending upon the input data. Once trained, the proposed method will remember the classification task and subsequently become able to classify the attack whenever a new set of inputs are provided to the pre-trained model. The proposed RNN and LSTM model has been tested over the NSL-KDD data-set and evaluated for varying different parameters (i.e., layers with a fixed number of neurons) and the width of the proposed models (i.e., number of neurons on a single hidden layer). The results proved that the proposed model performs better than the existing models in terms of accuracy.