Nonlinear dimensionality reduction for intrusion detection using auto-encoder bottleneck features

The continuous advances in technology is the reason of integration of our lives and information systems. Due to this fact the importance of security in these systems increases. Therefore, the application of intrusion detection systems as security solutions is increasing year by year. These systems (IDSs) are considered as a way of protection against cyber-attacks. However, handling big data constitutes one of the main challenges of intrusion detection systems and is the reason of low performance of these systems from the view of time and space complexity. To address these problems we have proposed an approach to reduce this complexity. Our approach is based on dimensionality reduction and the neural network bottleneck feature extraction is considered as the main method in this research. We have conducted several experiments on a benchmark dataset (NSL-KDD) to investigate the effectiveness of our approach. The results show that our approach is promising in terms of accuracy for real-world intrusion detection.