# Predictive Policing **Entity class:** Law-enforcement analytic approach **Maturity:** Developed ## Definition **Predictive policing** applies statistical or machine-learning models to forecast places, times, events, or persons judged at elevated risk. Systems trained on enforcement-generated data can reproduce patrol and arrest patterns, so predictions require provenance, validation, error analysis, and limits on consequential use. RAND's 2013 practitioner guide defines predictive policing as applying analytical techniques—particularly quantitative techniques—to identify likely targets for police intervention and prevent crime or help solve past crimes. Its taxonomy distinguishes methods for predicting crimes, future offenders, perpetrators of past crimes, and potential victims. RAND also states that predictive policing is not a crystal ball and identifies poor-quality data, misunderstood causal factors, weak evaluation, and overlooked privacy or civil rights as recurring pitfalls. ## Temporal inversion Predictive policing moves the analytic object upstream from a completed event toward the conditions from which an event might emerge: **crime → preparation → capability → network state → risk state → conditional future** This is a shift from [[wiki/Event-Centered Policing|event-centered policing]] toward [[wiki/State-Centered Policing|state-centered policing]]. It includes place forecasting, person-focused risk assessment, network analysis, anomalous-transition detection, and active collection driven by [[wiki/Value of Information|value of information]]. The output is a probability distribution, an uncertainty statement, and a ranked set of possible interventions. ## Predictive loop **Observe → Resolve Entities → Construct Relationships → Infer Hidden State → Quantify Uncertainty → Forecast Transitions → Identify the Most Informative Missing Observation → Lawfully Collect or Fuse Evidence → Update → Determine Whether Intervention Is Justified** This closed loop connects [[wiki/Entity Resolution|entity resolution]], [[wiki/Counterterrorism Predictive Graph|predictive graphs]], [[wiki/Inverse Problem|inverse problems]], [[wiki/Data Assimilation|data assimilation]], [[wiki/Uncertainty Quantification|uncertainty quantification]], and [[wiki/Collection Management|collection management]]. ## Civic boundary “Precrime” is the philosophical name for intervention before a forecast event. [[wiki/Probabilistic Precrime|Probabilistic precrime]] is its technical form. Observation, prioritization, inquiry, investigation, restriction, detention, and prosecution occupy different positions in the decision architecture and invoke different authorities. The system preserves [[wiki/Ground Truth|ground truth]], observed evidence, inferred state, and predicted state as different computational and institutional objects. ## Relationships - **method includes:** [[wiki/Risk Terrain Modeling|Risk Terrain Modeling]]. - **critiqued through:** [[wiki/Garbage In Gospel Out|Garbage In, Gospel Out]] and [[wiki/Predict and Surveil|Predict and Surveil]]. - **overlaps with:** [[wiki/Real-Time Observability|Real-Time Observability]] and [[wiki/Graph Analytics|Graph Analytics]]. - **temporal orientation:** [[wiki/Event-Centered Policing|Event-Centered Policing]], [[wiki/State-Centered Policing|State-Centered Policing]], and [[wiki/Pre-Event Inference|Pre-Event Inference]]. - **philosophical destination:** [[wiki/Probabilistic Precrime|Probabilistic Precrime]]. - **sensing substrate:** [[wiki/Federated Sensing Grid|Federated Sensing Grid]]. - **collection route:** [[wiki/Counterterrorism Collection - Predictive Policing and Probabilistic Precrime|Predictive Policing and Probabilistic Precrime]]. - **learning failure modes:** [[wiki/Runaway Predictive Feedback Loop|Runaway Predictive Feedback Loop]], [[wiki/Observation-Induced Ground Truth|Observation-Induced Ground Truth]], and [[wiki/Selective Labels|Selective Labels]]. - **harm accounting:** [[wiki/The Second Error Function|The Second Error Function]] and [[research/Harms Incurred While Bringing Preventive Systems Online|Harms Incurred While Bringing Preventive Systems Online]]. ## Sources / Provenance - [[research/A Network-Epidemiological Analysis of Radicalization and Real-Time Counterterrorism Observability|Network-Epidemiological Analysis]] - [RAND — Predictive Policing: The Role of Crime Forecasting in Law Enforcement Operations, 2013](https://www.rand.org/content/dam/rand/pubs/research_reports/RR200/RR233/RAND_RR233.pdf) - [GAO — Artificial Intelligence: Emerging Opportunities, Challenges, and Implications, March 28, 2018](https://www.gao.gov/products/gao-18-142sp) - [GAO — Law Enforcement: DHS Could Better Address Bias Risk and Enhance Privacy Protections for Technologies Used in Public, December 3, 2024](https://www.gao.gov/products/gao-25-107302)