An ARIMA-based forecasting system that analyses historical crime data to predict temporal patterns and trends, supporting more efficient allocation of policing resources across districts and time periods.
Type
Time Series Forecasting
Domain
Public Safety
Methods
ARIMA, Decomposition
Status
Completed
The Challenge
Law enforcement agencies allocate patrol and investigative resources based on historical patterns and operational experience, but these approaches struggle with changing trends, seasonal variations, and emerging hotspots.
Without predictive capability, resource allocation is inherently reactive, responding to crime after it occurs rather than positioning resources where they are most likely to be needed. Statistical forecasting can bridge this gap.
Approach
01
Data Analysis
Analysed Baltimore police department crime records, exploring temporal patterns by day, week, month, and year. Identified seasonal cycles, long-term trends, and district-level variations.
02
Time Series Modelling
Applied ARIMA models to the crime count data, with careful attention to stationarity testing, differencing, and parameter selection using AIC/BIC criteria.
03
Forecasting and Validation
Generated forward projections and validated against held-out test periods, assessing forecast reliability across different crime categories and geographic areas.
04
Actionable Reporting
Translated statistical forecasts into practical resource allocation recommendations, identifying peak-risk periods and districts.
CRIME PATTERN FORECASTING
SARIMA
Results
Temporal
Seasonal and trend patterns quantified
District
Geographic variation mapped
Forecast
Forward projections for resource planning
The forecasting models successfully captured seasonal crime patterns and longer-term trends, providing a statistical basis for forward-looking resource allocation. The district-level analysis revealed significant geographic variation in crime patterns, enabling more targeted deployment recommendations.