Loading Baltimore Police Department crime data spanning 2011-2015 and preprocessing for time series analysis.
Strategic division of time series data into training and test sets for robust model validation.
Autocorrelation Function (ACF) analysis to confirm seasonal patterns in crime data.
Specification of five distinct SARIMA models with varying parameter combinations.
Comprehensive fitting of all SARIMA models using maximum likelihood estimation.
Creating 52-week forecasts with confidence intervals for all trained models.
Comprehensive evaluation using Mean Squared Error (MSE) to identify optimal model.
Automated model selection using grid search with validation constraints.
Generating forecasts using the automatically selected optimal model.
Comprehensive evaluation determining Auto ARIMA as the superior approach.
ACF analysis revealed strong yearly seasonal patterns with peak correlation at lag 52 (0.6543), confirming predictable crime cycles that enable accurate forecasting for resource planning.
Auto ARIMA selected model SARIMA(1,0,2)(0,1,1,52) achieved MSE of 1.6675, outperforming the best manual model by 30.3%, demonstrating the power of automated model selection.
The optimal model incorporates both autoregressive and moving average components with seasonal differencing, capturing complex temporal dependencies in crime patterns.
Average forecast error of ±1.3 incidents per week with 95% confidence intervals provides reliable planning ranges for operational decision-making and resource allocation.
Comprehensive validation process with 36 model combinations tested and stability checks ensures robust model selection and prevents overfitting issues.
The analysis demonstrates that automated model selection significantly outperforms manual approaches, providing actionable insights for strategic policing decisions.