<p>The AnYa-type fuzzy set-based type-0 fuzzy rule was proposed in 2012 to address complex and high-dimensional data stream problems. Its core principle involves utilising the concept of data cloud (DC) to summarise internal data distribution when building fuzzy rules. Researchers preferred to use prototype instances to confirm the focal point of a DC and ensure the interpretability of fuzzy rules. However, prototype-based DCs typically require a large number of data instances to generate highly convincing DCs. Consequently, offline learning and chunk-by-chunk-based online learning are commonly utilised. In addition to prototype-based DC, the absence of mean-based DC results in a lack of clarity regarding the performance difference between prototype-based and mean-based online learning machines of DCs. Accordingly, this study firstly proposes a weighted online learning strategy to build local DCs based on their global DC. This approach enables training a classifier on a one-by-one data instance from a data stream without waiting for data chunks to arrive. Secondly, this study utilises small samples (insufficient training data instances) to reveal the performance difference between mean-based and prototype-based DCs. In the validating phase, a classic classifier structure, 0-order type-0 online fuzzy classifiers, is employed. Additionally, synthetic and real-world datasets are considered simultaneously. Results indicate their differences and disadvantages, demonstrating that the proposed weighted learning strategy of type-0 fuzzy DC holds promising potential for further development in various data mining technologies and practical problems.</p>