In this study, we propose a novel data stream clustering algorithm, called EdgeCluster, which has modest computational requirements, making it suitable for deployment on edge devices. The proposed EdgeCluster algorithm dynamically maintains the set of clusters that model the normal behavioral scenarios of a monitored streaming phenomenon by analyzing each new data portion and reflecting it in the clustering model. In addition, EdgeCluster is able to identify data objects that exhibit deviating behavior. The performance of the EdgeCluster algorithm is studied in several experimental scenarios using synthetic and real-world datasets. In these experimental scenarios, EdgeCluster is evaluated and compared to three other stream clustering algorithms, EvolveCluster, DenStream, and ClusTree, using several widely used cluster validation indices as well as a temporal SI measure, which is specifically designed for streaming data. The experimental results show that EdgeCluster, compared to other clustering algorithms, is able to handle overlapping and non-spherical clusters, and thus better captures real-world scenarios where the monitored phenomenon often exhibits several not clearly distinguishable operating modes.