A Novel Artificial Intelligence-Based Intrusion Detection System—NAI2DS

Intrusion Detection Systems (IDS) are traditionally known as an effective monitoring system and have been highlighted as an important layer in today’s ICT (Information and Communication Technology) environment due to the urgency of cyber security in everyday life. However, these IDSs face serious challenges due to the speed and variety of information and connected devices, which not only cause the continuous emergence of new attacks that are unknown to existing security systems, but the huge amount of data to be processed also exceeds the system’s potential. In addition, traditional detection methods generally protect users from cyber attacks after certain types of attack have appeared. For all these reasons, it is necessary to develop intelligent, scalable detection systems that are capable of detecting unknown attacks. This paper presents a new fast, lightweight and reliable intrusion detection system based on deep learning, composed of a multiphase methodology combining CNN (convolutional neural network) and BiLSTM (Bidirectional Long-Short Term Memory). The proposed methodology is validated on the CIC-IDS-2017 (Canadian Institute for Cybersecurity-IDS-2017) public dataset using a web attack. The results prove that the proposed methodology outperforms the baseline and existing approaches, achieving high classification performance, up to 99% accuracy. Our simulation results show that the proposed CNN-BiLSTM combination performs best compared to state-of-the-art work in the same dataset and based on various evaluation metrics.