Seattle Crime Analysis (2024)
MSIS 502 · Team 4: Yousef, Alysia, Shraddha, Trevor
Overview
A data-driven study of Seattle Police Department's 2024 crime data (81,771 reports) to identify top offenses, hotspot neighborhoods, and temporal trends. We also developed daily incident forecasts to guide staffing and prevention strategies.
Dataset
- Source: Seattle PD public crime records (filtered to 2024).
- Fields: offense category/subcategory, date/time, neighborhood, precinct/beat, lat/long.
- Cleaning: standardized timestamps, removed invalid rows, completed daily index.
Objectives
- Identify frequent crime types and subtypes.
- Locate geographic hotspots by neighborhood and precinct.
- Analyze temporal patterns and forecast daily incident volumes.
Key Insights
- Top category: Larceny/Theft (vehicle-related thefts prominent).
- Hotspots: Capitol Hill, Downtown, Queen Anne, Northgate, South Lake Union.
- Seasonality: Lower incidents in Feb–Mar; higher in Sep–Oct.
- Hourly/Weekly: Weekday peaks in late morning–early afternoon; lowest 12–5 a.m.
Modeling & Forecasting
Goal: Forecast daily citywide incidents (train Jan–Oct 2024, validate Nov–Dec 2024).
- Features: trend index, weekday dummies, annual seasonality (sin/cos), date continuity.
- Linear Regression: MAE 19.1 · RMSE 23.3 — best overall accuracy.
- Random Forest: MAE 25.31 · RMSE 31.14 — captured short-term spikes.
Takeaway: Simpler seasonal-trend features generalized best; richer multi-year data could further improve performance.
Recommendations
- Reinforce security in high-theft areas: parking access control, CCTV, lighting.
- Deploy targeted patrols at hotspot neighborhoods and peak hours.
- Enable real-time alerts and community awareness campaigns.
Challenges & Next Steps
- One-year window limits seasonality detection → extend to multi-year trends.
- Ambiguity between report vs offense timestamps → align with response-time data.
- Integrate socioeconomic and environmental covariates for improved forecasting.