Unveiling insights from unstructured wealth: a comparative analysis of clustering techniques on blockchain cryptocurrency data

Abstract

In the fourth industrial revolution era of today, individuals encounter an immense volume of information daily. The digital world is rich in data like IoT, social media, healthcare, business, cryptocurrencies, cybersecurity, etc. The situation can become problematic as these vast amounts of data require significant storage capacity, which leads to challenges in executing tasks such as analytical operations, processing operations, and retrieval operations that are time-consuming and arduous. To effectively analyze and utilize this data, artificial intelligence, particularly machine learning, and deep learning, can provide a practical solution. Clustering, an unsupervised learning technique, aims to identify a specific number of clusters to effectively categorize the data through data grouping. Hence, clustering is related to many fields and is used in various applications that deal with large datasets. This survey examines seven widely recognized clustering techniques, namely k -means, G -means, DBSCAN, Agglomerative hierarchical clustering, Two-stage density (DBSCAN and k -means) algorithm, Two-levels (DBSCAN and hierarchical) clustering algorithm, and Two-stage MeanShift and k -means clustering algorithm and compares them with a real dataset - The Blockchain dataset, including prominent cryptocurrencies like Binance, Bitcoin, Doge, and Ethereum, under several metrics such as silhouette coefficient, Calinski-Harabasz, Davies-Bouldin Index, time complexity, and entropy.