FASE Framework Aims to Reduce Bias in Predictive Policing
A team of researchers has introduced a novel framework called FASE, which stands for Fairness-Aware Spatiotemporal Event Graph, aimed at addressing racial biases in predictive policing practices. The framework combines crime prediction with equitable distribution of police patrols and features a feedback simulator for practical use. Focusing on Baltimore, the study analyzed nearly 140,000 serious crime cases reported from 2017 to 2019 across 25 ZIP Code Tabulation Areas. Utilizing advanced statistical methods, the researchers achieved significant results, indicating the framework's potential to mitigate biases and improve fairness in law enforcement resource allocation.
Key facts
- FASE stands for Fairness-Aware Spatiotemporal Event Graph framework
- The framework integrates crime prediction, fairness-constrained patrol allocation, and a closed-loop simulator
- Baltimore is modeled as a graph of 25 ZIP Code Tabulation Areas
- 139,982 Part 1 crime incidents from 2017 to 2019 were used
- Data is at hourly resolution
- Prediction uses a spatiotemporal graph neural network and multivariate Hawkes process
- Outputs use Zero-Inflated Negative Binomial distribution
- Validation loss: 0.4800, test loss: 0.4857
Entities
Locations
- Baltimore
- United States