Crime analysis using domain level similarity
Datasets relating time information to crime occurrences in the geographical regions are received. Time based crime patterns are extracted. Based on similarities among the crime patterns, the geographical regions are clustered. A selected time series dataset is augmented with a second time series dataset from the same cluster. Based on the augmented time series dataset, a new crime pattern is extracted. Based on the new crime pattern, a crime forecast is made for the selected geographical region.
1. A method of operating a crime forecasting system, comprising:
receiving a first time series dataset associated with a target geographical region, the first time series dataset relating time information to crime occurrences in the target geographical region;
calculating a first time based crime pattern based on the first time series dataset;
receiving clustering information that relates the target geographical region to a first set of substantially non-overlapping geographical regions, the first set of non-overlapping geographical regions including the target geographical region;
augmenting the first time series dataset with a second time series dataset to create an augmented time series dataset, the second time series dataset to be based on at least one time series dataset relating time information to crime occurrences in at least one of the first set of non-overlapping geographical regions that are not the target geographical region;
calculating a second time based crime pattern based on the augmented time series dataset; and
based on the second time based crime pattern, forecasting a crime pattern for the target geographical region,
wherein the at least one time series dataset relating time information to crime occurrences correspond to members of a subset of the first set of non-overlapping geographical regions that are not the target geographical region,
wherein the subset is selected based on at least one attribute associated with both the target geographical region and each of the subset,
wherein the at least one attribute is associated with at least one of population, demographics, economy, education, and land use,
wherein the at least one attribute is further associated with a proximity to at least one of a law enforcement facility, educational facility, and transportation facility, and
wherein the at least one attribute is predictive of a crime from the forecasted crime pattern.
2. The method of claim 1 , wherein the clustering information that relates the target geographical region to the first set of substantially non-overlapping geographical regions is based on statistical similarities between respective time based crime patterns associated with the first set of substantially non-overlapping geographical regions.
3. The method of claim 1 , wherein the clustering information that relates the target geographical region to the first set of substantially non-overlapping geographical regions is based on statistical differences between respective time based crime patterns associated with a second set of substantially non-overlapping geographical regions, the second set of substantially non-overlapping geographical regions not sharing any common members with the first set of substantially non-overlapping geographical regions.
4. A method of forecasting crime occurrences, comprising:
receiving a plurality of time series datasets that are each associated with crime occurrences in a respective geographical region;
receiving a set of attributes that are associated with each geographical region;
calculating respective statistical feature sets from each of the time series of datasets;
based on the statistical feature sets, associating the respective geographical regions with one of a plurality of clusters;
calculating statistical measures of independence between the attributes associated with the geographical regions and crime patterns associated with the geographical regions of the cluster; and
based on the statistical measures of independence, determining at least one crime predictive rule based on at least one statistical measure of independence meeting a threshold criteria,
wherein the at least one attribute is associated with at least one of population, demographics, economy, education, and land use, and
wherein the at least one attribute is further associated with a proximity to at least one of a law enforcement facility, educational facility, and transportation facility, and
wherein the at least one attribute is utilized in determining the at least one crime predictive rule.
5. The method of claim 4 , wherein the statistical feature sets correspond to patterns, in time series datasets, that relate crime occurrences to time information.
6. The method of claim 4 , wherein associating the geographical regions with one of a plurality of clusters is based on measurements of similarity between clusters as compared to similarity within clusters.
7. The method of claim 4 , wherein the attributes comprise demographic attributes and functionality attributes.
8. The method of claim 4 , wherein the statistical measures of independence relate a statistical dependence of crime occurrences in the geographical regions to the attributes associated with the geographical regions.
9. The method of claim 4 , wherein the respective statistical feature sets are based at least in part on a feature distribution.
10. The method of claim 9 , wherein the feature distribution may be selected from a set comprising one or more of gaussian, t, chi-square, poisson and inverse gaussian.
11. A crime pattern prediction system, comprising:
a network interface to receive a first time series dataset associated with a target geographical region, the first time series dataset relating time information to crime occurrences in the target geographical region, the network interface to also receive clustering information that relates the target geographical region to a first set of substantially non-overlapping geographical regions, the first set of non-overlapping geographical regions including the target geographical region;
a processor; and,
a non-transitory computer readable medium having instructions stored thereon that, when executed by the processor, at least instruct the processor to:
calculate a first time based crime pattern based on the first time series dataset;
augment the first time series dataset with a second time series dataset to create an augmented time series dataset, the second time series dataset to be based on at least one time series dataset relating time information to crime occurrences in at least one of the first set of non-overlapping geographical regions that are not the target geographical region;
calculate a second time based crime pattern based on the augmented time series dataset; and
based on the second time based crime pattern, forecast a crime pattern for the target geographical region,
wherein the at least one time series dataset relating time information to crime occurrences correspond to members of a subset of the first set of non-overlapping geographical regions that are not the target geographical region,
wherein the subset is selected based on at least one attribute associated with both the target geographical region and each of the subset,
wherein the at least one attribute is associated with at least one of population, demographics, economy, education, and land use,
wherein the at least one attribute is further associated with a proximity to at least one of a law enforcement facility, educational facility, and transportation facility, and
wherein the at least one attribute is predictive of a crime from the forecasted crime pattern.
12. The system of claim 11 , further comprising:
a display to present a forecasted crime pattern for the target geographical region.
13. The system of claim 12 , wherein the clustering information that relates the target geographical region to the first set of substantially non-overlapping geographical regions is based on statistical similarities between respective time based crime patterns associated with the first set of substantially non-overlapping geographical regions.