A STUDY ON APPLYING SOME DEEP LEARNING ALGORITHMS FOR EARLY NETWORK INTRUSION DETECTION

Abstract

This paper proposes and builds a model to evaluate the effectiveness of Deep Learning algorithms including Recurrent Neural Network (RNN), Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU), thereby determining the reliability of each dataset in building a network instrusion detection model. Because the models have similar structures, the evaluation will ensure objectivity. The results show that the algorithms applied on CICIDS2017 give a higher accuracy rate than the CSE-CICIDS2018 and the GRU model gives the best results. The study also shows that Deep Learning algorithms built on RNNs are relatively effective when it comes to detecting network attacks better than basic Machine Learning algorithms, which are capable of detecting a number of hidden features. both the CICIDS2017 and CSE-CICIDS2018 datasets are more reliable than the older ones.