An ensemble learning approach for intrusion detection in IoT-based smart cities

The increase in population is a huge threat to mankind and especially in cities, it is difficult to manage energy consumption, resource allocation, and maintaining security. This leads to large-scale urbanization and the formation of new cities with the integration of new technologies. The creation of smart cities could be a preferred choice to manage a huge population in a small area. This could be done with the usage of the Internet of Things (IoT) which employs different sensors to perform various tasks like data collection, traffic control, and weather detection. The data collected with these sensors are stored in the cloud and used for various applications. The large amount of data present in the IoT and its transparent nature attract various attackers. This leads to an increasing number of cyber-attacks in IoT, generating security issues to the data confidentiality for the individuals in the smart cities. Hence there is a need for the detection of these attacks commonly referred to as intrusions. The mechanism built to detect the intrusions is known as the Intrusion Detection System (IDS). Machine Learning (ML) is the commonly used technique in the creation of IDS as it shows superior performance in detection and classification works. This paper proposes an Ensemble Gradient Random forest-based Leopard Seal Search Optimization (EGR-LSS) algorithm. The hyperparameters of the proposed system are optimized using the Leopard Seal Search optimization algorithm. The comprehensive experiments are conducted to assess the proposed EGR-LSS model's detection efficacy utilizing the CICIDS2017 dataset. The proposed model outperformed the state-of-the-art techniques including CNN, AI-BC, and NB, and gained accuracy of 98.75%, recall of 97.4%, precision of 98.7%, and F1-score of 96.9% respectively. Overall, the proposed model provided reliable strong cyber threat detection performance and it increased the prediction speed by significantly reducing the risk prediction time.