Site Selection of Elderly Care Facilities Based on Multi-Source Spatial Big Data and Integrated Learning

Yin Zhang, Junhong Zhu, Fangyi Li, Yingjie WangISPRS Int. J. Geo Inf 2024 · DOI

TL;DR

This paper tackles the critical challenge of optimally locating elderly care facilities in aging regions by integrating multi-source spatial big data with advanced machine learning. By developing a novel prediction model and conducting multi-stage screening, the research provides a data-driven approach to ensure equitable and efficient distribution of these vital services.

What Problem Does This Paper Address?

As global populations age, particularly in countries like China, the demand for elderly care facilities is skyrocketing. However, the current approach to building these facilities is often ad-hoc, leading to suboptimal locations, uneven distribution, and ultimately, a potential decline in the quality of life for the elderly. The "15-minute elderly service circle" concept highlights the need for convenient access to care, but achieving this requires scientific planning that considers a multitude of factors. Existing research often relies on single data sources or less sophisticated methods, failing to fully leverage the power of big data and advanced analytics for this complex spatial optimization problem. This paper aims to bridge this gap by developing a robust, data-driven site selection model.

Methods

The study employs an integrated learning approach, combining multiple machine learning algorithms to enhance prediction accuracy. Specifically, it compares six classification algorithms: Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Classification and Regression Trees (CARTs), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT). The core of the methodology involves:

  1. Data Integration and Processing: Collecting and processing 26 types of multi-source spatial big data (including topographic, demographic, economic, transportation, and healthcare data) at a 500m grid scale. This data is then quantified and integrated into a comprehensive indicator system.
  2. Model Construction and Evaluation: Building an integrated learning model by training and comparing the performance of single learners and ensemble learners (RF and GBDT). A comprehensive evaluation using metrics like Accuracy, Precision, Recall, F1-score, and AUC is performed.
  3. Feature Importance Analysis: Identifying the most influential factors for site selection using RF and GBDT models.
  4. Site Selection and Screening: Generating preliminary suitability maps and then performing secondary screening based on population aging levels (7%, 14%, and 20% thresholds) to identify priority areas for new facilities.

Datasets

The study utilizes a variety of multi-source spatial big data for Hefei City, China. Key datasets include:

  • Terrain Conditions: Digital Elevation Model (DEM) data (GEBCO), used to derive altitude, slope, and plane curvature.
  • Economic Status: GDP spatial distribution grid data, average housing prices, and land transaction prices.
  • Transportation Accessibility: Road network data (OSM), river network data (OSM), and Points of Interest (POI) data for transportation facilities and automobile-related services (Amap).
  • Healthcare Facilities: POI data for healthcare facilities (Amap).
  • Living Convenience: POI data for dining, leisure, entertainment, sports, fitness, life services, science, and cultural education (Amap).
  • Population Distribution: Population distribution data (WorldPop) and census data for the elderly population.

No new benchmark datasets were introduced; rather, existing datasets were integrated and processed for the specific study area.

Key Insights and Ideas

  • Ensemble Learning Dominance: Integrated learning models (RF and GBDT) significantly outperform single learning models in predicting elderly care facility suitability, demonstrating their ability to handle complex spatial data.
  • Key Drivers of Suitability: The "closest distance to healthcare facilities" and "population distribution scale" are identified as the most critical factors influencing site selection. This underscores the importance of accessibility to medical care and the presence of a target demographic.
  • Layered Aging and Targeted Planning: The study highlights that different levels of population aging require differentiated site selection strategies. This nuanced approach moves beyond a one-size-fits-all solution.
  • Urban-Suburban Divide: Preliminary results show higher suitability in urban centers and lower suitability in suburban and rural areas, suggesting a need for targeted development strategies to address this disparity.

Key Results

  • Model Performance: RF and GBDT models achieved comprehensive evaluation indexes above 80%, outperforming single learning models.
  • Prediction Accuracy: The RF model showed an 87.9% fit with existing elderly facilities, while the GBDT model achieved a 78.4% fit, indicating the models' reliability.
  • Feature Importance: The top two factors influencing site selection were consistently identified as "closest distance to healthcare facilities" and "population distribution scale."
  • Secondary Screening Findings:
    • For an "aging society" (≥7% aging), high suitability areas are concentrated in main urban areas and some southern regions.
  • For a "deeply aging society" (≥14% aging), high suitability areas are primarily in main urban areas.
  • For a "super-aging society" (≥20% aging), high suitability areas are found on the peripheries of central urban areas and in specific super-aging townships.
  • Policy Recommendations: The study provides targeted recommendations for urban centers, suburban/peripheral areas, and rural areas, emphasizing infrastructure development, public-private partnerships, and regional equity.

Limitations and Future Work

  • Data Incompleteness: The authors acknowledge that the types of multi-source spatial big data used are not exhaustive and suggest incorporating more heterogeneous datasets in future research.
  • Algorithm Scope: While comparing six algorithms, the study suggests that exploring additional machine learning algorithms, such as neural networks, could further enhance prediction accuracy.
  • Dynamic Demand: The current model is based on static data. Future work could incorporate dynamic changes in population demographics and demand for elderly care services.
  • Broader Geographic Scope: While Hefei City is a valuable case study, applying and validating this model in other regions with diverse demographic and geographic characteristics would be beneficial.

Summary updated

BibTeX
@article{zhang2024site,
  author = {Yin Zhang and Junhong Zhu and Fangyi Li and Yingjie Wang},
  title = {{Site Selection of Elderly Care Facilities Based on Multi-Source Spatial Big Data and Integrated Learning}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2024},
  doi = {10.3390/ijgi13120451},
}
Figure