Crime in Philadelphia: Bayesian Clustering with Particle Optimization.

Area of Science:

• Spatial statistics
• Urban criminology
• Bayesian inference

Background:

• Accurate estimation of crime dynamics is crucial for urban public safety.
• Bayesian hierarchical modeling is suitable for analyzing neighborhood-level crime data.
• Standard models can over-smooth crime patterns due to spatial discontinuities.

Purpose of the Study:

• To develop a novel Bayesian prior to prevent over-smoothing in crime pattern analysis.
• To introduce an efficient ensemble optimization procedure for identifying neighborhood clusters.
• To improve the accuracy of crime trend estimation and forecasting in urban environments.

Main Methods:

• Developed a novel Bayesian prior that partitions neighborhoods into clusters to encourage within-cluster spatial smoothness.
• Introduced an ensemble optimization procedure with a new local search strategy to efficiently identify high-probability partitions.
• Applied the method to estimate crime trends in Philadelphia from 2006 to 2017.

Main Results:

• The proposed method demonstrated good performance in both estimation and partition selection on simulated and real data.
• The new prior effectively addresses over-smoothing issues caused by spatial discontinuities in crime patterns.
• The ensemble optimization procedure proved computationally efficient compared to conventional stochastic search techniques.

Conclusions:

• The developed Bayesian approach with a clustered prior and ensemble optimization enhances the accuracy of urban crime trend analysis.
• This method offers a more principled way to handle spatial heterogeneity in crime data.
• The findings have implications for improving public safety strategies in large cities.