IDDLE: A Novel Deep Learning-Based Approach for Intrusion Detection Problem Using Feature Extraction

The internet has revolutionized access to information, enabling users to retrieve vast amounts of data effortlessly, transcending geographical and temporal barriers. This unprecedented connectivity has advanced research, education, and communication, but it has also created vulnerabilities that cybercriminals exploit to steal confidential information, interfere with processes, and carry out extensive attacks. Consequently, robust security measures are essential to safeguard the integrity and confidentiality of online data. Intrusion detection systems (IDS) are crucial in defending against such threats, employing both signature-based and anomaly-based models. Even though signature-based IDS are highly effective at countering recognized threats, they struggle with novel, "zero-day" attacks. Conversely, anomaly-based detection systems can identify unknown threats but often generate high false positive rates. A hybrid IDS, combining elements of both approaches, offers a more comprehensive defense. This study presents a new intrusion detection model utilizing deep learning, evaluated on the KDD'99 and NSL-KDD datasets. The proposed model (IDDLE-intrusion detection deep learning engine) incorporates advanced preprocessing techniques, including normalization, feature extraction, and categorical encoding. Empirical findings demonstrate that the proposed model outperforms existing state-of-the-art approaches. This study underscores the capability of deep learning to enhance it in improving IDS performance, emphasizing the importance of continuous innovation and collaboration among security researchers, developers, and users to stay ahead of evolving cyberthreats. The findings underscore the significance of advanced feature extraction and hybrid detection strategies in developing robust intrusion detection systems.