Feed-forward neural network for Network Intrusion Detection

The Internet connection is becoming ubiquitous in embedded systems, making them potential victims of intrusion. While in the age of deep learning, these algorithms tend to produce worse results than traditional machine learning. In this paper, we propose a methodology based on feed-forward neural network for intrusion detection. Better performances than traditional machine learning techniques can be achieved when all steps of the methodology are applied. Performance is evaluated on CICIDS2017, showing accuracy better than 99% and a false positive rate lower than 0.5%. After analysis of previous studies, their results are compared to the performance of the proposed approach. Finally, the neural network trained on a PC has been implemented on an automotive processor to characterize performance aspects.