The Internet of Things (IoT) is springboarding novel applications and has led to the generation of massive amounts of data that can offer valuable insights across multiple domains: Smart Cities, environmental monitoring, healthcare etc. In particular, the availability of open IoT data streaming from heterogeneous sources constitute a novel powerful knowledge base. However, due to the inherent distributed, heterogeneous and open nature of such data, metadata that describe the data is generally lacking. This happens especially in contexts where IoT data is contributed by users via cloud-based open data platforms, in which even the information about the type of data measured is often missing. Since metadata is of paramount importance for data reuse, there is a need to develop intelligent techniques that can perform automatic annotation of heterogeneous IoT datastreams. In this paper, we propose two novel IoT datastream classification algorithms: CBOS and TKSE for the task of metadata annotation. We validate our proposed techniques through extensive experiments using public IoT datasets and comparing the outcomes with state-of-the-art classification methods. Results show that our techniques bring significant improvements to classification accuracy.