A Hybrid Deep Learning Framework for IoT Security Enhancement and Anomaly Detection

Cyber threats targeting Internet-of-Things(IoT) devices have grown in frequency and severity alongside the IoT's fast uptake and usage in the past few years. Because of this, creating a model that can detect hostile attacks and lessen security risks in IoT-devices is becoming a very important subject. Botnets pose a significant risk because they disrupt the networks that Internet of Things (IoT) devices rely on to function. An method called GA-HDLAD, which stands for genetic-algorithm with hybrid deep learning(DL), was created in this study to advance security in the IoT environment by detecting botnets. Botnet detection makes use of hybrid deep learning, which combines attention concepts, feature-extraction-techniques(FETs), and recurrent-neural-networks(RNNs). The hybrid-DL approach can identify botnet attacks, which frequently use complicated patterns. In addition, the model's RNNs capture temporal dependencies, while FETs guarantee efficient feature extraction from geographical data. To choose the hyperparameters required by the HDL method, simulated annealing (SA) is employed. This paper compares the GA-HDLAD system to other detection approaches and finds that it outperforms them experimentally using a benchmark botnet dataset.