Defining irregular clusters by using tree data structure and microclusters in real time streaming data problem

Streaming data clustering is a popular field for processing and extracting meaningful information from data streams in real-time. Streaming data clustering offers efficient approaches for defining and updating clusters based on similarity criteria. One of the significant challenges in streaming data clustering is identifying non- spherical clusters. However, outliers and noisy data in the streaming data clustering area can significantly degrade clustering performance. In contrast, methods capable of defining non-spherical clusters can enhance the clustering quality. As a solution to these problems, this study proposes a new streaming data clustering approach called MCMSTStream. The proposed algorithm combines the KD-Treebased micro-cluster and Minimum Spanning Tree (MST) based methods to achieve high clustering performance. Our proposed algorithm consists of five stages, which include defining KD-Tree-based microclusters, and then forming macro-clusters by applying MST to defined micro-clusters, defining new microclusters when new data arrives or deleting micro-clusters when the number of data falls below the threshold value of N, assigning new micro-clusters to macro-clusters or deleting macro-clusters when the number of micro-clusters falls below n_micro. As the last step, all system information is updated. The MCMSTStream algorithm that is proposed in this study is robust against outliers and noisy data and can detect arbitraryshaped clusters. Additionally, the proposed algorithm is fast and capable of handling high-dimensional data. In this study, the proposed algorithm's clustering success is measured using the Adjusted Rand Index (ARI) and Purity indices. To demonstrate the efficiency of our algorithm, we conducted experimental studies on state-of-the-art algorithms such as DenStream, DBSTREAM, and KD-AR Stream algorithms. The findings are presented in tables and graphs in detail. The results indicated that the proposed algorithm outperformed the compared algorithms. Furthermore, the time complexity of the algorithm is quite satisfactory.