Toronto Robbery Risk, 2020

Classification of Robbery Crimes in Urban Toronto

This project combines Toronto Police Service open incident records with neighbourhood census indicators to study how robbery risk shifts across time, place, and local socio-economic context. The analytical focus is not just where robberies happen, but where robberies account for an unusually large share of serious incidents.

16,611

serious crime incidents analyzed across Toronto in 2020

16.7%

share of analyzed Toronto incidents that were robberies in 2020

Moss Park

neighbourhood with the highest robbery count in this analysis

Background

Understanding the interplay of socio-economic, temporal, and spatial factors behind urban crime is necessary for making prevention strategies more targeted. In Toronto, robbery is especially relevant because it concentrates personal safety risk into a smaller set of places and time windows than many other serious crimes.

Research Question

Do temporal patterns, spatial context, and neighbourhood socio-economic conditions significantly affect the likelihood that a serious crime incident in Toronto is a robbery?

Variables of interest

  • Temporal factors: season, month, day of week, hour of day, and a daylight versus dark-hours indicator.
  • Spatial factors: neighbourhood, longitude, latitude, and premise type.
  • Socio-economic factors: individual income, household structure, rental concentration, labour-force stability, and related neighbourhood census indicators.

Data Sources

  • Toronto Police Service Public Safety Data Portal: auto theft, bicycle theft, break and enter, homicides, robbery, shooting and firearm discharge, and theft over.
  • Toronto Open Data neighbourhood profiles: 2021 census-derived socio-economic indicators for Toronto neighbourhoods.

Analytical Approach

  • Data wrangling: harmonize incident files, create temporal features, and merge crime incidents to neighbourhood census records.
  • Exploratory analysis: compare robbery share and robbery concentration across premise types, neighbourhoods, and time windows.
  • Predictive modeling: benchmark GLM, tree-based models, bagging, random forest, boosting, and XGBoost.

What the data shows

Neighbourhood concentration

Robbery burden is highly concentrated in a relatively small number of neighbourhoods, especially lower-income, rental-dense parts of the downtown east corridor and northwest Toronto.

Premise type dominates

Transit, outdoor, commercial, and educational settings carry far higher robbery shares than houses and apartments. This makes location setting one of the clearest signals in the project.

Hour beats darkness

The binary Darkness variable is only weakly informative overall. Hour-of-day and month produce clearer and more defensible patterns than a simple day versus night split.

The visualization page focuses on robbery share as well as robbery count, which helps separate high-activity settings from high-risk settings.

Modeling Summary

Random Forest and Bagging remain the strongest models in the existing modeling workflow, with Random Forest showing the best precision-recall balance among the implemented classifiers. The supporting interpretation also aligns with the EDA: spatial and contextual variables such as neighbourhood location, premise type, and hourly timing appear to carry more signal than broad seasonal heuristics alone.

References

[1] Stevens, H. R., Beggs, P. J., Graham, P. L., and Bi, P. “Hot and bothered? Associations between temperature and crime in Australia.”

[2] Baird, A., While, D., Flynn, S., Ibrahim, S., Kapur, N., Appleby, L., and Shaw, J. “Do homicide rates increase during weekends and national holidays?”

[3] Toronto Open Data Catalogue. https://open.toronto.ca/dataset/neighbourhood-profiles/

[4] Toronto Police Service Public Safety Data Portal. https://data.torontopolice.on.ca/pages/open-data