Blockchain technology has become a vital innovation in various aspects. Categorizing blockchain data streams has emerged as a popular area of studies. Several studies have been carried out in this field. Development of semi-supervised classification in conjunction with developing class identification is still a challenging task. In this paper, we proposed an effective and efficient blockchain data stream classification called Improved Very Fast Decision Tree (IVFDT). The IVFDT improves the traditional Very Fast Decision Tree algorithm (VFDT) by including adaptive splitting criteria and an efficient pruning mechanism that suit high-dimensional of blockchain data streams. The proposed approach tackles the issues of heterogeneity of the data, concept drift and the resources constraints and allows accurate and scalable classification in real time situations. Experiments on real-world blockchain datasets show that IVFDT outperforms other competing classifiers in accuracy and processing time.