Skip to main content

TPS OPEN DATA STUDY · TORONTO · 2014 TO 2025

The citywide total hides what changed

Toronto Auto Theft records rose 240% from 2014 to 2023, then declined. Assault remained the largest category from 2014 to 2025.

464,571Records analysed
12Years covered
5Categories
CATEGORY SHARESAll records
Assault
53.5%
Break and Enter
17.9%
Auto Theft
16.5%
Robbery
8.5%
Theft Over
3.5%

Share of analysed police-reported records, 2014 to 2025. Theft Over covers property theft above $5,000, excluding auto theft.

01 · START WITH THE ROW

464,571 records share 414,033 event IDs

Each row is an offence or victim record. Several rows can belong to the same police-reported event.

FOLLOW THE UNIT

One dataset, different counts

How the study counts records474,819 downloaded rows minus 10,248 partial-2026 rows leaves 464,571 analysed records. Those records contain 414,033 distinct event IDs. 456,337 records appear in the spatial grid. These are overlapping views, not additional records.474,819Downloaded rows2014 through partial 202610,248Excluded from analysisPartial report year: 2026464,571Records analysedComplete years: 2014–2025414,033Unique event IDsCount distinct IDs in the records456,337Records in the mapWithin the study’s spatial gridHow the study counts records474,819 downloaded rows minus 10,248 partial-2026 rows leaves 464,571 analysed records. Those records contain 414,033 distinct event IDs. 456,337 records appear in the spatial grid. These are overlapping views, not additional records.474,819Downloaded rows2014 through partial 2026Exclude 10,248partial-2026 rows464,571Records analysedComplete years: 2014–2025414,033Unique event IDsCount distinct IDs in the records456,337Records in the mapWithin the study’s spatial grid
The analysis keeps offence and victim rows. Counting distinct event IDs does not replace those rows in the charts. The map includes 98.2% of analysed records; 8,234 records are outside the mapped subset. The event-ID and map counts describe overlapping views and must not be added together.
Why exclude partial 2026?

An incomplete year has less observation time than a full year. Including it in the annual comparison would create an artificial decline. All analysis views here cover 2014–2025.

How to read the countA record is not a unique incident. Treating every row as a separate incident would overstate the number of events.

02 · Time

Auto Theft peaked in 2023; Assault peaked in 2024

Compare annual record counts across the five categories from 2014 to 2025.

20257,363 records

-23.8% from 2024

Annual reported records, 2014 to 2025Use left and right arrow keys on the chart to select a year and read its exact count.010,00020,00030,00040,0002014201620182020202220242025Auto Theft · 2023 · 12,592Assault · 2024 · 25,806

What the total leaves outAuto Theft rose 240% from 2014 to 2023 and then declined. Assault remained the largest category every year, but followed a different pattern.

03 · Category patterns

Auto Theft records peak at 22:00

Choose a category to compare how its own records are distributed by hour, day of the week, and location type.

More than half of Auto Theft records were associated with outside locations. The hourly distribution peaked at 22:00.

Time of day

24-hour profile

Peak 22:00 · 8.8%

Share of records by occurrence hourEach point is the percentage of this category recorded in one hour of the day.0%3%6%9%12%00:0006:0012:0018:0023:00

Share within Auto Theft records with a recorded occurrence hour · fixed 0–12% scale.

Day of week

Weekday profile

Shared 0 to 18% scale

Share within Auto Theft records with a recorded weekday. Each column starts at zero.

Context

Location types

Within-category share · 0–100%

Apartment3.3%
House30.4%
Commercial8.3%
Outside54.5%
Transit0.3%
Educational0.1%
Other3.0%

Share within records with a recorded premises type; the location order stays fixed.

Two examplesAuto Theft peaks at 22:00, and most records are linked to outside locations. Break and Enter has its largest weekday share on Friday, with houses and apartments accounting for more than half of its records.

04 · Space

All categories combined, West Humber-Clairville has the most records

Compare the five neighbourhoods with the most records from 2014 to 2025. Counts are not adjusted for population or activity.

Neighbourhood aggregation

Most records in this view

  1. 1
    West Humber-Clairville12,734
  2. 2
    Moss Park10,247
  3. 3
    Downtown Yonge East9,363
  4. 4
    York University Heights9,105
  5. 5
    Yonge-Bay Corridor8,862

Raw record counts, not population-adjusted rates.

Ranking excludes records without a named neighbourhood. The grid separately includes records inside the coordinate bounds; the two subsets can differ.

Explore the coordinate grid
Approximate coordinates

Records by coordinate grid

456,337 records inside the coordinate bounds

Aggregated spatial density in TorontoA privacy-preserving grid of approximate coordinates. Darker cells indicate greater density on a shared logarithmic count scale across all categories.
1 record10,354 records

Shared logarithmic count scale. This coordinate grid has no geographic basemap; cells are not neighbourhoods.

What the map cannot showNeighbourhood counts have no denominator for population, visitors, vehicles, traffic, or land use. A higher count does not establish a higher risk for an individual.

05 · Model audit

Random forest accuracy: 59.2% versus a 58.2% baseline

The model classifies an existing record from its context. It does not forecast whether or where a future event will happen.

CHRONOLOGICAL SPLIT

Earlier records train the model. 2025 tests it.

Training, validation and test yearsTraining: 2014–2023, 374,136 records. Validation: 2024, 47,402 records. Test: 2025, 43,033 records. Scores below use the 2025 chronological test set.201420152016201720182019202020212022202320242025Training374,136 records10 complete yearsValidate47,402recordsTest43,033recordsTraining, validation and test yearsTraining: 2014–2023, 374,136 records. Validation: 2024, 47,402 records. Test: 2025, 43,033 records. Scores below use the 2025 chronological test set.Training · 2014–2023374,136 recordsValidation · 202447,402 recordsTest · 202543,033 records
Training, validation and test records come from successive report years. Every score below refers to the same 43,033 records in 2025.
2025 chronological test

Accuracy

Accuracy on the 2025 test setMajority baseline: 58.2%. Logistic regression: 44.4%. Random forest: 59.2%. All bars start at zero and share a full zero-to-one scale.0%25%50%75%100%Majority baseline58.2%Logistic regression44.4%Random forest59.2%
+0.99 percentage pointsRandom forest versus the majority baseline

Accuracy counts all correct predictions. The majority baseline predicts Assault for every record.

The random forest correctly classified 59.2% of test records, about 1 percentage point above always predicting Assault (58.2%). The accuracy gain was small.

2025 chronological test

Macro F1

Macro F1 on the 2025 test setMajority baseline: 0.147. Logistic regression: 0.368. Random forest: 0.424. All bars start at zero and share a full zero-to-one scale.0.000.250.500.751.00Majority baseline0.147Logistic regression0.368Random forest0.424
+0.277Random forest versus the majority baseline · Macro F1

Macro F1 averages the five category F1 scores equally. Higher is better; the scale runs from 0 to 1.

The random forest scored higher than the Assault-only baseline on Macro F1 because it also recognised records from smaller categories.

View full normalized matrix
Rows are true categories; columns are model predictions. Each row uses that true category’s test records as its denominator. Stronger shading means larger shares; bold cells are correct predictions. Rounded rows may not sum to exactly 100%.
True \ PredictedAssaultBreak and EnterAuto TheftRobberyTheft Over
Assault64%14%12%8%2%
Break and Enter24%60%11%2%2%
Auto Theft14%8%68%7%2%
Robbery35%13%23%24%5%
Theft Over32%25%25%9%10%

Inspect one true category

True category

Robbery

Precision17.8%

Recall24.5%

F10.206

2,566 records in the 2025 test set
Predicted as

Assault35%

Break and Enter13%

Auto Theft23%

Robbery24%

Theft Over5%

Reading the category scoresPrecision is the share of predicted labels that are correct; its denominator is all predictions of that category. Recall is the share of actual category records found, matching the matrix diagonal. F1 combines precision and recall.

06 · Study scope

Scope at a glance

Police-reported records across five categories, 2014–2025.

01

Records

Offence or victim rows; multiple rows can share an event ID.

02

Neighbourhood counts

Raw totals, without adjustment for population or activity.

03

Model task

Classify existing records, without forecasting future events.

RESEARCH MATERIALS

Read the methods or reproduce the analysis

The paper documents the full methodology. The code package includes a 5,000-row stratified sample and the outputs used on this page.

Share this study