머신러닝 기반 지방소멸 요인의 구조 변화와 대응 전략:삶의 질 요인을 중심으로

Abstract

This study investigates the impact of multidimensional Quality of Life (QoL) indicators on regional extinction in South Korea, moving beyond traditional economic-centric approaches. Panel data encompassing 24 QoL indicators from 228 municipalities (2015–2023) were analyzed using a convergent methodology. Fixed effects panel regression was employed to control for macroscopic causalities, combined with XGBoost and SHAP analysis to identify feature importance and non-linear interactions. Results identified the proportion of aged housing, the number of cultural facilities, and the number of elderly welfare facilities as the three core determinants. Time-series SHAP analysis revealed a structural shift: the dominant influence of physical infrastructure (e.g., aged housing) has gradually decreased, whereas the relative importance of cultural and welfare indicators has continuously risen. This confirms that the underlying causes of regional extinction are shifting toward qualitative dimensions. Furthermore, K-means clustering based on SHAP vectors classified municipalities into three distinct types. This study contributes a bottom-up regional typology model reflecting unique local vulnerabilities. It suggests that to effectively overcome regional extinction, the policy paradigm must shift from uniform job provision toward customized strategies focused on building family-oriented settlement conditions where housing, welfare, and culture are organically integrated.