| 摘要: |
| 步行对老年人的身心健康至关重要。在
积极应对人口老龄化和健康中国两大战略背景
下，构建满足老年人步行出行需求、支持老年人
步行活动的空间，已成为城乡规划学科的一项
紧迫任务。现有研究已广泛证实了社区建成环
境与老年人步行行为的密切关联，但大多未能充
分考虑变量之间可能存在的非线性关系及其协
同效应。本研究以成都市为例，借助POI和街景
图像等多源大数据，运用前沿的可解释机器学习
方法（融合LightGBM和SHAP模型），分析了
社区建成环境与老年人步行频次之间的非线性
关系和变量之间的协同效应。结果表明，对老年
人步行频次影响最大的3个建成环境变量是人
行道占比、归一化植被指数和休闲娱乐设施可
达性。SHAP模型进一步揭示了社区建成环境对老年人步行行为存在阈值效应，并详细描述了变量之间的协同效应。最后，提出了完善社区服
务设施建设、提高社区步行道通达性、营造高品质公共空间，以及考虑建成环境要素的交互作用
四个方面的规划建议。本研究为城乡规划学科参与老年人活动研究展现了新的视角，并为以支持
步行活动为导向的适老步行环境规划设计提供了科学支撑。 |
| 关键词: 建成环境 物质环境 步行行为 人口老龄化 健康老龄化 可解释机器学习 |
| DOI：10.13791/j.cnki.hsfwest.20240611 |
| 分类号: |
| 基金项目:国家自然科学基金面上项目（52278080）；成都市哲
学社会科学规划一般项目（2023BS128）；唐仲英基
金会包容性城市规划建设联合研究平台项目（2022
009）；成都市哲学社会科学研究基地—成都公园城
市示范区建设研究中心项目（GYCS2023-YB005） |
|
| Non-linear and synergistic effects of built environment factors on older people’s walkingfrequency in Chengdu: A Shapley additive explanations analysis |
|
WEI Dong,YANG Linchuan
|
| Abstract: |
| Population aging is a global trend. To address this issue, the World Health Organization (WHO)
has introduced the concepts of “active aging” and “healthy aging”, emphasizing the importance
of maintaining health and well-being throughout the aging process. China has also placed high
importance on the aging issue, and has proposed the national strategy of “Actively Coping with
Population Aging.” In this context, the health problems of older people have received unprecedented
attention. However, older people often face various health challenges, due in part to insufficient
physical activity. Moderate physical activity can decrease the occurrence of coronary heart disease,
Type-2 diabetes, and hypertension among older people. It also alleviates anxious and depressive
symptoms, enhancing their quality of life.
Walking, as a fundamental form of physical activity, is crucial for the health of older people.
It helps maintain physical functions and promotes mental health and social interaction. However, in
reality, to accommodate growing traffic demand, modern urban development has reduced walking
spaces and sidewalks. This reduction further creates obstacles for older people to engage in walking.
In the context of population aging, it is crucial to optimize the urban walking environment to better
accommodate and encourage the walking behavior of older people.
The built environment refers to the physical spaces shaped by human activities in urban or rural
areas. It includes buildings, streets, transportation facilities, public spaces, etc. Existing research has
extensively confirmed that various factors of the built environment significantly impact the walking
behavior of older people. Key factors include population density, land use mix, road network density,
street intersections, and street greenery.
Most existing research employs traditional analytical methods that are predicated on assumed
generalized linear relationships. However, these methods face significant limitations, as they fail
to capture the non-linear effects of the built environment on the walking behavior of older people.
To address this limitation, researchers have recently turned to “black-box” machine learning
models, such as random forests, gradient boosting decision trees (GBDT), extreme gradient
boosting (XGBoost), and light gradient boosting machine (LightGBM). These advanced models
can reveal non-linear relationships and threshold effects, thereby leading to widespread application.
Despite these advancements, existing research still has the following shortcomings: (1) difficulty in
explaining the decision-making process of “black-box” machine learning models; and (2) neglect of
the interactions between built environment factors.
The Shapley additive explanations (SHAP) model is a powerful tool that helps us understandhow machine learning models make predictions by analyzing the impact of each feature on the prediction results. This feature enhances our understanding
of the machine learning model’s decision-making process. The significance of SHAP lies in its ability to explain the output of any machine learning
model. Therefore, this study applies cutting-edge interpretable machine learning methods by integrating LightGBM and the SHAP model to address the
aforementioned research gaps. It utilizes multi-source big data, including points of interest (POIs), remote sensing images, and street view images, to analyze
the non-linear relationships between the community-level built environment and the walking behavior (specifically, daily walking frequency) of older people
in Chengdu. Furthermore, this study reveals the synergistic effects between built environment factors.
The key findings of this study are as follows: (1) The three most important built environment factors affecting the walking frequency of older people
are the proportion of sidewalks, the normalized difference vegetation index (NDVI), and accessibility to recreational facilities. (2) There are significant nonlinear
and threshold effects of built environment factors on older people’s walking frequency. For example, when the proportion of sidewalks and the NDVI
are within the ranges of 0.01~0.04 and 0.06~0.12, respectively, they have a positive contribution to the predicted output (older people’s walking frequency)
compared to the model's baseline prediction, the average prediction across the dataset. The relationship between the green view index and the walking
frequency of older people follows an inverted U-shape: when the green view index is less than 0.23, they are positively correlated; when the green view index
exceeds 0.23, they are negatively correlated. (3) Significant interactions, which may reveal where optimizing one feature is more impactful in the presence of
another, exist between many built environment factors. Notable synergistic effects exist between the green view index and the floor area ratio, the land use
mix and the sky view index, and accessibility to recreational facilities and the floor area ratio. Specifically, when the green view index is greater than 0.25
and the floor area ratio exceeds 1.50, the land use mix exceeds 0.40 and the sky view index exceeds 0.30, and accessibility to recreational facilities exceeds 23
and the floor area ratio exceeds 1.50, the SHAP interaction value of the two variables is greater than 0, which indicates that the two variables enhance older
people's walking frequency more than expected based on the effects of each feature alone.
This study has significant practical implications. Firstly, the threshold effect indicates that independent variables have an optimal impact on the walking
behavior of older people within a specific range. Secondly, promoting walking behavior is more effective when modifying a set of built environment factors
rather than changing a single factor. Third, this study proposes planning strategies from the perspectives of improving the construction of community service
facilities, enhancing the accessibility of community walkways, and creating high-quality public spaces. |
| Key words: built environment physical environment walking behavior population aging healthy aging explainable machine learning |
