Machine-Learning Technique for Drafting Regional Population Policy: Cluster-Based Demographic Typology of Rural Municipalities in the Three Baltic Countries

Aleksandrs Dahs, University of Latvia
Juris Krumins , University of Latvia

To ensure implementation of effective and quickly adaptable population policies in the context of diverse demographic situation and limited resources, researchers and policy makers must develop and improve effective tools for differentiation of territorial units, which would allow dividing them into typical groups according to their socio-demographic risks and development potential. This study aims to develop and test a machine-learning technique for categorization of territorial units based on their demographic characteristics that can be applied in drafting regional population policy measures and monitoring their performance over time. Proposed methodology relies on unsupervised non-hierarchical partitioning clustering algorithm. The study focuses on rural municipalities of the three Baltic countries – Estonia, Latvia and Lithuania, which represent diverse sample of regional demographic development modalities. Provided examples demonstrate that with sufficient data input, unsupervised machine-learning tools can be beneficial for drafting and monitoring regional population policy measures. Algorithms like PAM clustering can be used for efficient classification of territorial units according to their demographic characteristics. Using such an approach for smaller datasets (e.g., national, or regional level), provides more sensitive results even with a smaller number of clusters. This study shows that suggested technique provides informative and actionable results useful for policy planners.

See extended abstract

 Presented in Session 119. Flash session Policy Development and Measurement