Design of a Framework for Detecting Attacks in Internet of Things

Abstract

IoT-Fog computing provides a diverse array of services for end-based IoT systems, newlinewhere end devices communicate with both cloud and fog nodes to manage the client tasks. The newlineescalating occurrences of data breaches and cyberattacks exploit security vulnerabilities in IoT newlinedevices across various industries. IoT end devices are particularly susceptible to critical attacks, newlineincluding DDoS, and other security threats during data collection among the cloud and fog newlinelayers. Moreover, early detection of these network susceptibilities is crucial. DL plays a pivotal newlinerole in predicting end-user behavior through retrieving features and identifying threats in the newlinenetwork. However, the limited computational as well as storage capabilities of IoT devices newlinehinder the execution of DL on them. To address this limitation, this thesis proposes two novel newlineattack detection models in IoT-Fog computing. The first contribution introduces a Deep Hybrid newlineDetection Model for Attack Detection in IoT-Fog Architecture with three-stage process. It newlinebegins with improved Z-score normalization for data preprocessing, followed by feature newlineextraction (IG, Entropy, raw data, and improved MI) in the second stage, and the introduction newlineof a hybrid attack detection scheme using optimized DeepMaxout and DBN classifiers in the newlinethird stage. To optimize the training of DeepMaxout, a hybrid optimization model called newlineBMUJFO algorithm is introduced. The second contribution addresses both attack detection as newlinewell as mitigation in the network. Initially, the class imbalance problem is addressed through newlineadvanced class imbalanced processing. Then, handcrafted features such as improved entropy newlinebased features, correlation-based features, statistical features, and raw data are extracted to newlineprovide additional information related to attack behavior. newline