머신러닝을 활용한 농촌 빈집 발생 예측 및 격자기반 관리 방안 연구: 경상북도 상주시를 사례로 = Machine Learning-Based Prediction of Rural Vacant Houses and Grid-Based Management Strategies : A Case Study of Sangju-si, Gyeongsangbuk-do

다국어 초록 (Multilingual Abstract)

This study estimates the probability of housing vacancy occurrence in rural Sangju and develops differentiated management strategies by operationalizing these probabilities into grid-based spatial information. Drawing on 2022 vacant house survey data, a comparative analysis of five machine learning models identified XGBoost as the most robust predictor. SHAP-based interpretations further revealed that small housing size, older building age, and an aging population are the primary drivers of vacancy. Furthermore, the study proposes a four-tiered management framework - Intensive Intervention, Current Management, Preventive Management, and Monitoring - providing a systematic basis for policy prioritization. Ultimately, this research illustrates that machine learning–driven spatial diagnostics can serve as a practical foundation for proactive and evidence-based rural vacancy management, demonstrating how empirical diagnostics and policy design can be organically integrated.