The Internet of Things (IoT) is revolutionizing how electronic devices and sensors connect and communicate with each other, and it is helping industries and different sectors by providing real-time, data-driven insights. As suggested by Gartner’s study, currently, approximately 25 billion devices and gadgets are part of the IoT ecosystem, including wearables, smart homes, automated vehicles, and smart city applications ( Mishra and Tyagi in Artificial Intelligence-based Internet of Things Systems. Internet of Things. Springer, Cham, 2022). However, the nature of IoT devices also makes them prone to a wide range of security attacks and vulnerabilities, including anomalies. IoT devices are resource-constrained and do not have enough memory and computational power, so there is a need for capable machines that can be used to deal with a high computational load for machine learning purposes (Joshi and Korah in 2022 IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems (MASS), Denver, CO, USA, pp. 748–753, 2022) Anomaly detection in Internet of Things (IoT) networks is a critical aspect of ensuring the security, efficiency, and reliability of these networks. IoT networks consist of interconnected devices that communicate and exchange data, often with minimal human intervention. This paper delves into the realm of anomaly detection within the Internet of Things (IoT) networks, proposing a novel two-fold approach that combines cloud-based machine learning training with transfer learning to gateways. Transfer learning to gateways enhances real-time detection capabilities by leveraging pre-trained models, optimizing resource usage, and reducing latency. This paper aims to provide an in-depth analysis of machine learning-based anomaly detection methods aimed at enhancing the security of IoT networks.