Deep Learning Model with Normalized Bayesian Optimizer for Anomaly Classification in IoT Security

The Internet of Things (IoT) stands as a pivotal facilitator of technology, garnering considerable attention from the global scientific community. Nevertheless, the proliferation of IoT devices and the vast amounts of data they accumulate pose a significant vulnerability to an array of security threats and susceptibilities. The growing adoption of IoT infrastructure has given rise to challenges, including node failures, heightened threats, and increased susceptibility to attacks, anomalies, and potential security breaches. Addressing and mitigating these issues constitute a pivotal domain within the overarching realm of IoT. The current study presents a novel approach for anomaly classification in IoT Security by employing a Convolutional Neural Network (CNN) with feature optimization via the Normalized Bayesian Optimization Algorithm (NBOA). This research strives to enhance anomaly detection accuracy beyond the limitations of conventional CNN models. The proposed CNN is trained using meticulously optimized features extracted from the extensive IoT23 dataset, which is provided in CSV format. The utilization of this dataset results in notably superior performance, contributing to the overall effectiveness of the proposed anomaly detection approach. The proposed model attains an impressive accuracy of 98.69%, outperforming standard CNN models. The methodology entails leveraging NBOA for fine-tuning feature selection, thereby augmenting the model's ability to discern anomalies based on input values.