Abstract: This article is devoted to the research of economic growth of countries by identifying patterns in historical data sets on macroeconomic indicators. Using machine learning techniques, namely cluster analysis methodology in combination with data transformation algorithms, in particular dimensionality reduction, groups of countries with similar patterns in the structure of the economy, availability of production factors, internal and external economic activity and development dynamics were formed. The novelty of the article is the approach to selecting optimal clustering and dimensionality reduction algorithms by quantifying the results of their work. The evaluation of the dimensionality reduction methods was carried out using the cumulative variance indicator, and the clustering methods were assessed based on the aggregate indicator proposed in the article, which combines the standardized Davies-Bouldin, Calinski-Harabasz indices and the Silhouette coefficient. According to calculations, among the 11 considered methods of dimensionality reduction, the most effective is the Kernel PCA algorithm, while among the 7 clustering methods, K-means is the most effective for this task with a given set of indicators. The study was conducted on 6 five-year time intervals from 1991 to 2020 with a focus on the Ukrainian economy. According to the research, Ukraine’s economy migrated from the “post-Soviet” cluster (first half of the 1990s) to the Eastern European cluster (second half of the 2010s) over the period under consideration, which indicates real economic growth and gradual integration with the European Union.