Explainable Machine Learning Classification Of Rural Rescue Service Proximity Coverage Using Openstreetmap Data

Abstract
Assessing the spatial proximity of rural settlements to rescue-service locations remains an important task for sustainable community safety planning, particularly in regions characterized by dispersed settlements, limited infrastructure, and incomplete spatial data. This study proposes a data-driven framework for classifying rural rescue-service proximity-based coverage using OpenStreetMap-derived geospatial data and explainable machine learning methods. A spatially structured dataset containing 304 rural settlements was constructed using infrastructural and settlement-related indicators, including schools, medical facilities, places of worship, industrial facilities, buildings, and major-road access. The target variable represented binary proximity-based coverage defined using a 5 km Euclidean buffer around rescue-service locations and was intended solely for comparative territorial screening; it does not represent observed emergency-response time or operational performance. To ensure methodological validity, leakage-prone variables directly related to target construction were excluded from model training. Several classification approaches were comparatively evaluated using stratified cross-validation, including Logistic Regression, Random Forest, Gradient Boosting, MLP, and XGBoost models. The obtained results demonstrated that ensemble machine learning methods provided the most stable and interpretable performance for spatial tabular data. The Gradient Boosting model achieved the highest mean F1-score of 0.77, with a balanced accuracy of 0.85, ROC–AUC of 0.95, and PR–AUC of 0.82 under stratified five-fold cross-validation. Model interpretability was additionally assessed using permutation feature importance, SHAP analysis, and multicollinearity diagnostics, which identified model-specific associations between infrastructural characteristics and the proximity-based coverage class. Spatial decision-support maps were developed to identify settlements with lower model-based proximity-coverage probabilities that may warrant further detailed assessment. The proposed framework demonstrates the feasibility of integrating OpenStreetMap data and explainable machine learning into practical decision-support systems for sustainable rural safety management and infrastructure planning under conditions of limited and fragmented spatial information.

Keywords
OpenStreetMap
Proximity-based coverage
Rural communities
Explainable machine learning
Spatial decision support
Sustainable community safety
Geospatial analysis
Gradient boosting
SHAP analysis
Emergency planning

Cite this article
Tryhuba, A., Koval, L., Rudynets, M. et al. Explainable machine learning classification of rural rescue service proximity coverage using OpenStreetMap data. Sci Rep (2026). https://doi.org/10.1038/s41598-026-72396-z

Kontakt

Ukraiński Uniwersytet w Europie – Fundacja
Balicka 116, 30-149 Kraków, Polska
NIP: 6772498871, REGON: 526358177

Middle East and North Africa
Dr. Faten Mohammed Al Najjar
fatena_mn@yahoo.com
+961 71 901 827

Lithuania
Prof. Savelii Kukharets
savelii.kukharets@vdu.lt
+37 066 768 478

E-Mail

university.uue@gmail.com

Phone

+48 500 006 634
+48 669 383 659

Explainable Machine Learning Classification Of Rural Rescue Service Proximity Coverage Using Openstreetmap Data

Abstract
Assessing the spatial proximity of rural settlements to rescue-service locations remains an important task for sustainable community safety planning, particularly in regions characterized by dispersed settlements, limited infrastructure, and incomplete spatial data. This study proposes a data-driven framework for classifying rural rescue-service proximity-based coverage using OpenStreetMap-derived geospatial data and explainable machine learning methods. A spatially structured dataset containing 304 rural settlements was constructed using infrastructural and settlement-related indicators, including schools, medical facilities, places of worship, industrial facilities, buildings, and major-road access. The target variable represented binary proximity-based coverage defined using a 5 km Euclidean buffer around rescue-service locations and was intended solely for comparative territorial screening; it does not represent observed emergency-response time or operational performance. To ensure methodological validity, leakage-prone variables directly related to target construction were excluded from model training. Several classification approaches were comparatively evaluated using stratified cross-validation, including Logistic Regression, Random Forest, Gradient Boosting, MLP, and XGBoost models. The obtained results demonstrated that ensemble machine learning methods provided the most stable and interpretable performance for spatial tabular data. The Gradient Boosting model achieved the highest mean F1-score of 0.77, with a balanced accuracy of 0.85, ROC–AUC of 0.95, and PR–AUC of 0.82 under stratified five-fold cross-validation. Model interpretability was additionally assessed using permutation feature importance, SHAP analysis, and multicollinearity diagnostics, which identified model-specific associations between infrastructural characteristics and the proximity-based coverage class. Spatial decision-support maps were developed to identify settlements with lower model-based proximity-coverage probabilities that may warrant further detailed assessment. The proposed framework demonstrates the feasibility of integrating OpenStreetMap data and explainable machine learning into practical decision-support systems for sustainable rural safety management and infrastructure planning under conditions of limited and fragmented spatial information.

Keywords
OpenStreetMap
Proximity-based coverage
Rural communities
Explainable machine learning
Spatial decision support
Sustainable community safety
Geospatial analysis
Gradient boosting
SHAP analysis
Emergency planning

Cite this article
Tryhuba, A., Koval, L., Rudynets, M. et al. Explainable machine learning classification of rural rescue service proximity coverage using OpenStreetMap data. Sci Rep (2026). https://doi.org/10.1038/s41598-026-72396-z

More posts