REVIEW AND SELECTION OF CLUSTERING ALGORITHMS FOR DATASETS IN THE CONTEXT OF COUNTRIES' DECARBONIZATION

Abstract

The problem of assessing and modeling a country's decarbonization potential is crucial for ensuring sustainable economic and social development at the micro and macro levels. Given the increasing relevance of Big Data in decarbonization research and the integration of clustering algorithms, it is crucial to identify clustering methods that are scalable, robust and suitable for used dataset.  This paper is structured as follows: the initial stage includes a comprehensive review and analysis of the relevant literature. The results of the literature review suggest that no single optimal clustering method, so a comparative approach adapted to the nature of the dataset may provide the best results. The next stage provides an overview of widely used clustering methodologies applied to the prepared datasets, particularly in the context of decarbonization. The author also provided an assessment of the quality of clustering using internal clustering metrics.  The study has been performed on a pretrained dataset of 14 normalized key indicators to determine the decarbonization potential of 41 countries over a 10-year period.  Finally, the application of the three clustering methods (K-means, GMM and SOM) was tested on a database to assess the decarbonization potential of different countries, including Ukraine and important conclusions were drawn. The study concludes that SOM with 3 and 5 clusters is most suitable clustering for a dataset used to determine the decarbonization potential of countries, in particulate Ukraine. The obtained clustering results can be used to adapt best international practices to the Ukrainian energy infrastructure, which is undergoing adverse significant transformations as a result of the war.