Enhanced IoT Anomaly Detection Using Combined Machine and Deep Learning Techniques on the IoT-23 Dataset

The Internet of Things (IoT) has revolutionized daily life and business operations by providing enhanced accuracy, efficiency, and speed. However, IoT systems face a growing risk of cyber threats, including malicious software and denial-of-service (DoS) attacks, which compromise the integrity and security of data. Despite the widespread adoption of IoT, many implementations lack robust cybersecurity measures, leaving them exposed to various threats. This highlights the urgent need to develop more resilient and effective defenses for IoT assets. This paper focuses on implementing anomaly detection for IoT systems utilizing the IoT-23 dataset, a comprehensive dataset containing both benign and malicious network traffic from IoT based devices. The study evaluates the performance of several machine learning (ML) and deep learning (DL) models, including decision trees, the Extreme Gradient Boosting (XGBoost) model, the Naïve Bayes model, and fully connected neural networks (FCNN).Our findings demonstrate that decision trees accomplished the highest accuracy of 98.9% while also requiring the shortest processing time, making them the most efficient model for anomaly detection in IoT systems. In contrast, the Naïve Bayes model performed the poorest, achieving an accuracy of only 50%. These results emphasize the significance of choosing the appropriate algorithm to enhance IoT security, ensuring effective anomaly detection and improved resilience against cyber threats.