Authors: Thi-Thu-Huong Le, Jihyun Kim, Jaehyun Kim, Howon Kim

Journal: Journal of Platform Technology (JPT) 

Abstract: The Intrusion Detection System (IDS) is an effective method to deal with the types of problem in networks. In particular, IDS is an effective way to obtain high security in detecting intrusion activities. An anomaly detection is one of intrusion detection with a high false alarm, a moderate accuracy and a detection rates when it is unable to detect all types of attacks correctly. To address this problem, we propose new approach through deep learning field. By applying Gated Recurrent Unit (GRU) Recurrent Neural Network (RNN), we build an effective intrusion detection classifier with higher performance on the dataset is extracted from KDD Cup 1999 dataset. From our experiments, we obtain 97.06%, 98.65%, and 10.01% for detection rate, accuracy, and false alarm rate respectively. Our result shows that the proposed approach model outperforms other model using FNNN, GNNN, RBNN, KNN, SVM, etc. for accuracy, detection rate with a significant false alarm rate.

Link: Analyzing Effective of Activation Functions on Recurrent Neural Networks for Intrusion Detection -Journal of Multimedia Information System | Korea Science