Abstract: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.

Contextual Optimizer through Neighborhood Estimation for prescriptive analysis

Fast Rates for Contextual Linear Optimization

Addressing misspecification in contextual optimization

Robust Data-driven Prescriptiveness Optimization

Solving Contextual Stochastic Optimization Problems through Contextual Distribution Estimation

Conformal Contextual Robust Optimization

Contextual Optimization under Covariate Shift: A Robust Approach by Intersecting Wasserstein Balls

Online Contextual Decision-Making with a Smart Predict-then-Optimize Method

Online Personalized Assortment Optimization with High-Dimensional Customer Contextual Data

A Survey of Contextual Optimization Methods for Decision Making under Uncertainty

Efficient Real-world Testing of Causal Decision Making via Bayesian Experimental Design for Contextual Optimisation

Heteroscedasticity-aware residuals-based contextual stochastic optimization

Predict-then-Calibrate: A New Perspective of Robust Contextual LP

Context-Dependent Ranking and Selection under a Bayesian Framework

Sequential Probability Assignment with Contexts: Minimax Regret, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood

Observing Context Improves Disparity Estimation when Race is Unobserved

CoS: Enhancing Personalization and Mitigating Bias with Context Steering

Optimizing the Spatial Scale for Neighborhood Environment Characteristics Using Fine-Grained Data.

Selecting Optimal Decisions Via Distributionally Robust Nearest-Neighbor Regression.

Learning Relevant Contextual Variables Within Bayesian Optimization

Learning context-aware region similarity with effective spatial normalization over Point-of-Interest data