4.7 Article

Optimizing the spatial scale for neighborhood environment characteristics using fine-grained data

出版社

ELSEVIER
DOI: 10.1016/j.jag.2021.102659

关键词

Neighborhood effect; Multiscale feature; Spatial scale sensitivity; Mobile phone user data

资金

  1. Xinjiang Production and Construction Corps, China [2017DB005]
  2. National Foundation of Natural Sciences of China [42171327]

向作者/读者索取更多资源

The neighborhood effect of environment characteristics is an important topic in social sciences, but defining a neighborhood and its aggregation scales are challenges. This study proposes a multiscale feature extraction and scale optimization framework to determine the optimal scale for each variable and build an optimized multiscale model. The framework includes methods for multiscale extraction and a random forest-based scale optimization method. The research demonstrates the effectiveness of the proposed framework in urban burglary risk modeling.
The neighborhood effect of environment characteristics has been an important topic in social sciences that are concerned with people's behaviors. However, defining a neighborhood and its aggregation scales are difficult challenges for the studies using neighborhood characteristics. Previous studies generally adopted fixed administrative units as the neighborhood and chose the same scale for all the variables due to data availability. Aggregation scales have a significant influence on examining the relationships between area-based input variables and outcome variables, but how to optimize the scale for each variable has been less investigated. This study proposes a multiscale feature extraction and scale optimization framework for simultaneously determining the optimal scale for each neighborhood environment variable and building an optimized multiscale model. First, we developed multiscale extraction methods for the socioeconomic and built environment characteristics within overlapping neighborhoods based on different types of data, including fine-grained mobile phone user data and point of interest data. Next, we proposed a random forest-based scale optimization method to quantify the impact of scales and determined an optimal spatial scale for each variable. Third, we used urban burglary risk modeling as an example, to demonstrate the usefulness of the scale optimization method. In the comparison, the model using an optimized multiscale feature set (R2 = 0.843, F1 score = 0.797 for high-risk samples) achieved better accuracy than all the models using fixed-scale combination feature sets. Moreover, the results revealed that there might be differences in the optimal scale of different neighborhood environment characteristics and demonstrated the effectiveness of our scale optimization method. Our multiscale feature extraction and scale optimization framework is well suited for assessing the neighborhood effect in social sciences.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.7
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据