A Minority-Aware Deep Learning Framework for Intrusion Detection under Imbalanced Network Traffic

Abstract

As cyber assaults are on the rise, there is a requirement for a robust and dynamic IDS that efficiently detects harmful activity from network data. However, the class imbalance problem in the network data is a great challenge to the standard IDS. To overcome this problem, FS approaches are used, such as RFE, to minimise the dimensionality of the data and improve the performance of the intrusion detection models. In this paper, a new strategy based on an ensemble method is proposed which leverages the benefits of several classifiers to overcome the issue of class imbalance and improve detection accuracy. The model is tested with the CIC-IDS 2017 dataset which consists of different network traffic scenarios and attack patterns. With the advanced preprocessing and FS methods, the ensemble model has a good performance. Especially, the voting classifier (RF + DT) obtains an excellent accuracy of 99.5%. Based on the results obtained it is shown that the proposed technique is able to solve the problem of imbalanced data without sacrificing high detection performance and therefore is a promising technique to enhance the reliability and efficiency of IDS in real world cyber defence scenario.