DL‐IDS: a deep learning–based intrusion detection framework for securing IoT

• Published in: Transactions on emerging telecommunications technologies Vol. 33; no. 3
• Main Authors: Otoum, Yazan, Liu, Dandan, Nayak, Amiya
• Format: Journal Article
• Language: English
• Published: 01.03.2022
• DOI: 10.1002/ett.3803
• ISSN: 2161-3915

The Internet of Things (IoT) is comprised of numerous devices connected through wired or wireless networks, including sensors and actuators. Recently, the number of IoT applications has increased dramatically, including smart homes, vehicular ad hoc network (VANETs), health care, smart cities, and wearables. As reported in IHS Markit (see https://technology.ihs.com), the number of connected devices is projected to jump from approximately 27 billion in 2017 to 125 billion in 2030, an average annual increment of 12%. Security is a critical issue in today's IoT field because of the nature of the architecture, the types of devices, different methods of communication (mainly wireless), and the volume of data being transmitted over the network. Security becomes even more important as the number of devices connected to the IoT increases. To overcome the challenges of securing IoT devices, we propose a new deep learning–based intrusion detection system (DL‐IDS) to detect security threats in IoT environments. There are many IDSs in the literature, but they lack optimal features learning and data set management, which are significant issues that affect the accuracy of attack detection. Our proposed module combines the spider monkey optimization (SMO) algorithm and the stacked‐deep polynomial network (SDPN) to achieve optimal detection recognition; SMO selects the optimal features in the data sets and SDPN classifies the data as normal or anomalies. The types of anomalies detected by DL‐IDS include denial of service (DoS), user‐to‐root (U2R) attack, probe attack, and remote‐to‐local (R2L) attack. Extensive analysis indicates that the proposed DL‐IDS achieves better performance in terms of accuracy, precision, recall, and F‐score. The paper proposes a novel DL‐IDS to identify severe anomalies. An SMO algorithm is used to extract the most relevant features from the data set. An SDPN is then applied to identify the optimal features and classify the data as normal or anomalous in different attack categories (eg, DoS, U2R, R2L, and probe). DL‐IDS system achieved superior results; in accuracy (99.02%), precision (99.38%), recall (98.91%), and F1‐score (99.14%) using the NSL‐;KDD data set.