【目的】在快速老龄化背景下，养老服务设施的选址直接关系到可达性、公平性与资源配置效率，是智慧城市空间治理的关键环节。为降低主观经验对规划的依赖，构建一套可解释、可复用的选址判别框架，为城市养老设施的时序化建设与优化布局提供依据的任务日益紧迫。【方法】本文以南京市为研究案例，整合第七次全国人口普查数据高德地图2024年12月POI数据，在500 m网格尺度刻画现有养老设施与其周边设施构成特征，采用基于信息增益的ID3决策树学习现状区位的选址规则与标准，量化不同设施类型对养老设施空间格局的解释力。之后从南京市27269个网格中识别出4511个适宜建设养老服务设施的区位。进一步叠加各地区老龄化程度与人口密度，对这些候选区位进行二次筛选和街道级排序，得到更加精细化的养老设施选址方案。【结果】模型与现状布局比对的总体准确率为83.93%，验证了方法的有效性与可迁移性。特征重要性排序显示：日常生活服务设施为一级主导因子，公共管理与公共服务设施为二级因子，企业分布与文体休闲设施为三级因子，明确了影响现有养老设施空间格局的关键驱动要素。基于决策树规则的预测布局呈“核心—次边缘—边缘”的优化结构：江宁等主城外围地区应作为新增供给的优先布局带；浦口、高淳等区域可结合本地资源条件适度发展商业化养老项目，以满足多层次需求。新方案有效覆盖郊外围老年人口集聚区，缓解中心城区与外围之间的供给失衡。【结论】以现有设施为基础，用可解释的机器学习方法学习其选址规则，本文为养老设施选址提供可落地的时序化建设清单，降低规划主观性、提升决策一致性。该框架对其他公共服务设施的空间布局优化具有推广价值，为智慧城市与空间规划提供数据驱动的决策支持。
 

[Objective] In the context of rapid population ageing, the siting of elderly-care facilities directly affects accessibility, equity, and resource-allocation efficiency, and is a key component of smart-city spatial governance. To reduce reliance on subjective judgement, an interpretable and reusable siting-discrimination framework is needed to support phased construction and optimized layouts. [Methods] For Nanjing, the Seventh National Population Census and AMap point of interests (POI) (December 2024) were integrated. At the 500-m grid scale, existing elderly-care facilities and surrounding POI composition are characterized. An information-gain–based ID3 decision tree learns siting rules and criteria embedded in the current location pattern and quantifies the explanatory power of different facility types. From 27,269 grids, 4,511 suitable locations are identified. Ageing level and population density are then overlaid for second-stage screening and subdistrict-level ranking to produce a refined siting scheme. [Results] Overall predictive accuracy against the current layout reaches 83.93%, demonstrating effectiveness and transferability. Feature-importance ranking indicates daily-life service facilities as primary drivers, public administration and public service facilities as secondary factors, and enterprise distribution together with culture–sports–recreation facilities as tertiary factors shaping the existing pattern. The predicted plan exhibits a “core–semi-periphery–periphery” optimization structure: peri-urban belts such as Jiangning are prioritized for new supply, while Pukou and Gaochun can use local resource endowments to develop market-oriented elderly-care projects to meet diversified demand. Coverage of suburban concentrations of older adults is improved, alleviating supply imbalances between the urban core and outer areas. [Conclusions] Building on the existing facility system, an interpretable machine-learning approach learns siting rules and delivers an actionable, phased construction list, thereby reducing planning subjectivity and enhancing decision consistency. The framework is generalizable to the spatial layout optimization of other public-service facilities and provides data-driven support for smart-city spatial planning.