Deep Neural Network based Method for Detection and Classification of the Malicious Network Traffic

Abstract

Internet of Things (IoT) has revealed as a prominent technology in recent years that cause massive growth in network traffic. At the same time, intrusion threats have become more sophisticated. Many Intrusion Detection Systems (IDS) developed to monitor and find abnormal activities through network traffic. Due to huge network traffic and different types of attacks inspecting each packet is time consuming and computationally intensive. Therefore, stronger IDS are required to classify network traffic with the best performance. Similarly, data classification has become a great challenging task. Many esteemed Machine Learning (ML) techniques (supervised and unsupervised) are used to overcome this situation. Deep Learning (DL) techniques are implemented for packet inspection and attack identification. In this thesis, we have proposed IDS based on DNN to monitor traffic coming from authentic and non-authentic sources and significantly classify malicious traffic with the highest accuracy of 99.89% and minimum loss function 0.00063. We have applied sequential model while improving the number of hidden layers. The hybrid activation functions were applied namely sigmoid, ReLU, and Softmax. We have applied binary cross entropy loss function and achieved minimum loss of 0.00063 %. We have used KDD 99 dataset and applied feature extraction steps. With all the adjustments we achieved highest accuracy as compared to that present in literature.