The diverse and ever-expanding nature of IoT devices has led to new challenges in network security management. Traditional measures of network security usually are not enough in the IoT settings, hence requiring innovative methods for ensuring a safe environment. The security flaws need to be overcome if the IoT architecture is to be viable. For decades, various scholars within the industry and academia have been proposing ways of detecting and preventing computer security breaches. This paper proposes yet another different approach that involves using machine learning algorithm(s) in predicting node's behavior by looking at how the nodes behave while connected within the IoT network. The methodology proposed in this paper involves training models on the UNSW-NB15 dataset to distinguish between normal device activities and malicious ones. Random Forest Classification is selected as, compared to other methods like support vector machines and K nearest neighbors, it has higher accuracy and definition in detecting anomalous behaviors. The method also aggregates cosine similarity measures to show improvement in the detection of subtle anomalies, which will ensure that slight deviations from normal behaviors are detected. Results illustrate the potential of ML methods to enhance the reliability and security of the IoT ecosystem and be one step toward smarter IoT with more secure network management.