Spatial crime distribution and prediction for sporting events using social media

Figures & data

Table 1. Reported crime records in Chicago, Illinois, US Chicago police department’s CLEAR (counts for the five bins used in this study) *relative total includes just the seven crime types considered in this study; the real value includes 30 crime types.

Crime Type	Frequency
Assault	6,446 (8.32%)
Battery	18,875 (24.35%)
Criminal damage	11,010 (14.20%)
Motor vehicle theft	4,837 (6.24%)
Other offense	6,978 (9.00%)
Robbery	4,046 (5.22%)
Theft	25,313 (32.66%)
Relative Total*	77,505 (100%)

Table 2. Day selection for the five bins.

	Bulls		Blackhawks
	home	away	home	away	Control days
Mon-Thu	20	19	17	20	19
Fri-Sun	10	11	13	10	11

Figure 1. Feature selection using random forest.

Figure 2. Data used in this case study.

Table 3. Summary of predictor variables used in this analysis.

Crime history variables:

Historical crime data before prediction day for the five bins

Demographic variables:

Population at crime risk (different for the five bins), residential population; population white, population black or African American, population Asian, population 62 years and over, foreign-born (%), 25 years and over high school or General Educational Development, total 25 years and over, 25 years and over less than high school, 25 years and over some college, foreign-born, household with individuals under 18 years, population 18 years and over total, households by type: non-family, households by type: husband-wife family, Bachelor’s or higher studies (%), 25 years and over bachelor’s degree or higher, Hispanic or Latino (of any race), average household size of occupied housing units by tenure: owner-occupied, average household size of occupied housing units by tenure: renter-occupied, median age by sex for both sexes

Socio-economic variables:

Vacant housing units, homeowner vacancy rate (%), unemployed, households below poverty (%), below the poverty level (%), rental vacancy rate (%), occupied housing units, hardship index, income per capita, the price per person for Airbnb, Airbnb locations

Environmental variables:

restaurants, bars, bus stops, buildings, bike racks, transportation routes: density

stadium: distance

Dynamic variables:

Geo-located Twitter data for the five bins: density and distance

Violent Tweets for the five bins: density and distance

Figure 3. Prediction models design.

Figure 4. Example of surveillance plots for other offense for two different prediction days, home games Chicago Bulls.

Figure 5. Density distribution of crimes around the venue, where gray squares represent areas with similar crime densities, and brown square with higher crime density during game days; red dots circle is the 1km buffer around the venue.

Figure 6. Density distribution of geo-located tweets around the venue; red dots circle is the 1km buffer around the venue.

Figure 7. Density distribution of violent tweets around the venue; red dots circle is the 1km buffer around the venue.

Figure 8. Moran’s I index for aggregated and disaggregated crime types, tweets and violent tweets; unit of analysis is the city of Chicago and a 1km buffer around united center.

Figure 9. The sliding window prediction approach.

Figure 10. AUC and AUC improvement for the seven crime types *AUC real values are presented only for assault in order to save space.

Figure 11. AUC and AUC improvement for the five bins *AUC real values are presented only for home games Chicago bulls in order to save space.