This document summarizes a study that used data mining techniques to predict crime using real-world crime datasets from Denver and Los Angeles. The goals were to identify crime hotspots and predict future crime types based on location, time, and other attributes. The models tested included the Apriori algorithm to identify frequent crime patterns, a naïve Bayesian classifier to predict crime type based on location/time features, and a decision tree classifier. Key results identified crime hotspots and showed the Bayesian classifier achieved prediction accuracies of 51-54% while the decision tree was more complex and achieved lower accuracy.

1. Crime Prediction Using Data Mining
Alia ALHamwi
Joelle Shannakian
Khaled Ismaeel
Mohammad AlBash

2. Contents
• Introduction
• DataSet
• Data Processing
• Models Building
• Key Results

3. Introduction
• find spatial and temporal criminal
hotspots using a set of real-world
datasets of crimes.
• predict what type of crime might
occur next in a specific location
within a particular time.

4. Goal
• police forces & resources
• stay away
• improve living place choices

5. Data Sets
• Main Fields
• Crime Category
• DateTime
• Neighborhoods of the city (location)
• Is it Crime or accident ?

6. DENVER

7. Los Angeles

8. Data Processing
• Data Cleaning
• Data Reduction
– among the available 19 attributes in
Denver crimes dataset, we just selected four.
– crime vs accidents .
• Data Integration
– Crime_Type,Crime_Date, and Crime_Location .
– Crime_Date = Crime_Month, Crime_Day, and
Crime_Time .

9. Data Processing
• Data Transformation and Discretization

10. Models Building
• Apriori Algorithm
– It scans the dataset to collect all itemsets that
satisfy a predefined minimum support
(optimum choice  Denver 0.0012 & Los
Angeles 0.0018).
– List of all crime hotspots.
– Location and time features, without the crime
type feature.
– Restriction, three specific itemsets  Location,
Day, Time.

11. • Naïve Bayesian Classifier
– Supervised learning algorithm.
– Statistical model that predicts class membership
probabilities based on Bayes’ theorem.
– P (CrimeType|Location,Day,Month,Time)
– Crime features contain  month, day, time,
location.
– Class Label  Crime type.

12. • Decision Tree Classifier
– Supervised Learning algorithm.
– Learning simple decision rules implied from the
data features  predict class label.
– Best tree  entropy & Information gain.
– generated tree was complex  restriction  ten
maximum leaf nodes.

13. EVALUATION
• Apriori Algorithm
– frequent crime patterns
– )+) readiness , easiness ,implement
– (-) very long running time

14. EVALUATION
• Naïve Bayesian and decision tree
classifiers
– crime type prediction
– NBC: 51% Denver- 54% for LA
– DTC: 42% Denver- 43% for LA
– very complex tree
– tree that cannot generalize

16. Key Results
• Crime Frequent Hotspots
• Crime Prediction
• Crime Hotspots Demographics Analysis

17. Crime Frequent Hotspots
• Apriori algorithm
• finding spatial and temporal criminal
hotspots
• Denver : 62 patterns
• Los Angeles : 59 patterns.

18. Crime Prediction
• Bayesian classifier
– Month, Day of the week, Time, Location
– All features in nominal values

20. Crime Hotspots Demographics
Analysis
• Relationship between crime rate and
demographics
– large population and large number of housing
– vacant houses and the dangerous locations
– people’s age and gender
– More man  more dangerous
– More female  more safe
– ages from 20 to 29

23. References
• https://arxiv.org/ftp/arxiv/papers/1
508/1508.02050.pdf