Modified LSTM Based Intrusion Detection System for Networks

The proliferation of wireless networks in the past decade is seriously coupled with security threats and intrusion incidents, raising greater demands for robust and efficient intrusion detection systems. A new approach is present in this paper, to intrusion detection in wireless networks based on the computational paradigm Echo State Networks. ESN is one type of reservoir computing model. The inherent advantages of training efficiency and dynamic responses for temporal patterns make ESN especially fit for the task of detecting network traffic anomalies. In the following, we design our proposed ESN-based IDS to identify and classify various types of network intrusions with high accuracy and low computational overhead. This system will be robust and adaptive against different attack scenarios by training with the entire dataset of normal and malicious network activities. We show the efficiency of our technique through extensive simulations and real experimentation for a wide range of intrusion detection such as exploit as well as unauthorized access attempts. The results explicitly show that ESN-based IDS outperforms detection accuracy, false positive rate, and time response by traditional machine learning methods.