Data Mining University Crime

1
Data Mining University Crime

• Presenters Stuart King
• Konrad Kuczynski
• Ying Liu

2
Data Sources

• a) FBI Uniform Crime Reporting campus crime
  statistics for 669 schools from 1995 thru 2007
  (except 2004)
• 1995-2005  http//www.securityoncampus.org/crimes
  tats/index.html
• 2006-2007 
• http//www.fbi.gov/ucr/cius2006/offenses
  /standard_links/universities_colleges.html  
• http//www.fbi.gov/ucr/cius2007/offenses
  /standard_links/universities_colleges.html 
• b) City Crime Data
• 1999-2007 http//www.fbi.gov/ucr/ucr.htm
• c) State Poverty
• 1999-2007 http//www.census.gov/hhes/www/saipe/co
  unty.html

3
Data Mining Tasks

• Cluster/Classify
• 1 Prediction of missing population values
• Used for improving both charting and anomaly
  detection of Per Capita crime data
• 2 Prediction of crime trends
• Can be used by universities to allocate
  resources for the coming year
• Outlier Detection
• 3 Detect anomalies in crime
• Can be used by universities to target root
  cause and prevention projects

4
Data Preprocessing

• University crime data cleanup
• ? 669 universities with 971 names
• ? 136 universities were removed because they
  reported less than 5 years of data
• ? If a year of crime data was missing, a
  fabricated record of averages was added
• Matching City, State, and Zip code information
  for each University
• Post cleaned and merged data sets
• ? 5,869 records of crime data for 533
  universities with City/State/Zip code added
• ? 1,712 records of crime data for 249
  universities with the following added
• City/State/Zip, City Crime Data, State Poverty
  Data
• Additionally, Per Capita and ? values were
  calculated

5
Data Mining

• Data used for each task

St. StPov Pop. Vio. Murd. Rape Rob Aslt Prpty Burg Theft Car Arson
1 G
1 P
2 G U? U? U? U? U? U? U? U? U?
2 P S? U? C? U? C? U? C? U? C? U? C? U? C? U? C? U? C? U? C? U? C? U? C?
3 3 U? U? U? U? U? U? U? U? U?

• 1 Population cluster/classify
• 2 Crime Trend cluster/classify
• 3 Anomaly Detection
• G Grouping Clustering
• P Prediction Classifying

• Per capita values
• ? Difference values
• Absolute values
• U University
• C City
• S State

6
Data Mining

• Algorithm used for each task

Task Mining Algorithm
1 Population Prediction Clustering EM
1 Population Prediction Classification Decorate using J48
2 Crime Trend Prediction Clustering EM
2 Crime Trend Prediction Classification J48
3 Crime Anomalies Outlier Detection DBSCAN with minPoints1
7
Visualizations

• Summary charts and graphs for Per Capita and
  Clustering data
• Interactive Map showing cluster changes for 533
  Universities
• Interactive Map showing predicted clusters for
  249 Universities
• Interactive Map showing where 355 outliers
  occurred
• Interactive Charts showing values for outliers
• http//www.cse.msu.edu/kingstua/Team3