1. Atlanta's Increasing Crime
Atlanta has been increasing in crime despite strained policing resources. By optimizing police station locations, we hope to reduce crime rates and increase response efficiency. Previous research focuses on locating police stations close to crime hotspots but do not give details such as the staff composition. As different neighborhoods may have different types of crime, we aim to provide specific composition recommendations for each police station.
2. Crime, Crash, & Demographics Dataset
A total of 101 Atlanta neighborhoods were analyzed using the following datasets.
| Crime Data | Crash Data | Demographics Data | |
|---|---|---|---|
| Duration | 2015-2019 | 2015-2019 | 2019 |
| # Records | 136,175 | 204,336 | 103 |
| # Columns | 10 | 22 | 20 |
| Source | APD | ATL DOT | DataNexus |
3. Approach & Results
Cluster
We reduced the number of variables by choosing those with standard deviation > 0.01, then we used K means to cluster the neighborhoods using these factors. The elbow method was used to find the optimal K=4, then the neighborhoods of the same cluster were grouped into police zones.
Optimize
We created a mixed integer linear program to minimize the distance between the potential police stations (i) to crime locations. The candidate locations were generated by taking the intersection of vertical and horizontal lines plotted one mile apart from each other over each zone. The crime locations were divided into violent (j) and non-violent crime types (k), and a weight of 0.7 and 0.3 were given to each type, respectively, so that more importance is given to minimizing distance to violent crime type locations.
The binary variable \( x_i \) is 1 if potential location \( i \) is chosen and 0 otherwise. The parameter \( D_{H_{ij}} \) represents the distance in miles from location \( i \) to \( j \), and \( D_{L_{jk}} \) is the distance in miles from location \( j \) to \( k \). The parameter \( n \) is the number of police stations needed for the zone.
4. Evaluation
The binary variable xi is 1 if potential location i is chosen and 0 otherwise. The parameter DHij represents the distance in miles from location i to j, and DLjk is the distance is miles from location j to k. The parameter n is the number of police stations needed for the zone.