Internet of Things (IoT) systems are experiencing accelerated growth, providing multiple services in a wide variety of environments. However, this diversity also poses significant challenges in terms of security. The proliferation of threats such as Mirai or Dark Nexus malware clearly reflects the increase in attacks targeting IoT infrastructures. Currently, one of the most widely used protocols in the application layer of these networks is the Constrained Application Protocol (CoAP), which has vulnerabilities that can be exploited through denial of service (DoS) attacks. In this scenario, this paper analyses and compares the performance of four supervised classification techniques for the real-time detection of DoS attacks in IoT networks that use the CoAP protocol. For its validation, a dataset has been used that collects network traffic from an IoT environment affected by this type of attack. The best results were obtained using artificial neural networks, achieving an F1-Score above 0.99 and outperforming the other techniques evaluated.