Machine Learning in the Comprehensive Study of the Human Development Index

Abstract

Introduction: An urgent task of regional policy consists not only in monitoring the human development index, but also in identifying key socio-economic factors determining its level. Realizing these dependencies necessitates the development of evidence-based measures to reduce inter-regional inequality and improve the quality of life.Methods: The study is based on the analysis of official statistical data. The target indicator was transformed into a binary variable for classification tasks based on the median value. After processing the gaps and generating new features, six machine learning algorithms, including linear and ensemble methods, were trained and optimized. A stacking ensemble was additionally developed to integrate their strengths. The quality of the models was assessed using complex metrics in a test sample.Results: Correlation analysis has revealed a strong negative relationship between the index and the level of poverty and unemployment, and a strong positive relationship with the indicators of demographic burden and economic activity. Among the models, gradient boosting demonstrated the best predictive ability, reaching an accuracy of 0.8824 and a ROC-AUC value of 0.9523. The stacking ensemble has enabled us to obtain a balanced classifier with competitive indicators.Discussion: The results have confi rmed the effectiveness of modern ensemble methods for modeling complex socio-economic dependencies. The constructed models serve as a tool for analytical support of management decisions, providing the opportunity to assess the potential impact of changes in individual areas on the integral indicator of the region’s development. The future work will focus on providing a more profound interpretation of the models to develop specifi c recommendations.