Article ID: | iaor20082640 |
Country: | Netherlands |
Volume: | 43 |
Issue: | 1 |
Start Page Number: | 199 |
End Page Number: | 210 |
Publication Date: | Feb 2007 |
Journal: | Decision Support Systems |
Authors: | Chen Hsinchun, Kaza Siddharth, Wang Yuan |
Keywords: | artificial intelligence: expert systems |
In recent years border safety has been identified as a critical part of homeland security. The Department of Homeland Security searches vehicles entering the country for drugs and other contraband. Customs and Border Protection (CBP) agents believe that such vehicles operate in groups and if the criminal links of one vehicle are known then their border crossing patterns can be used to identify other partner vehicles. We perform this association analysis by using mutual information (MI) to identify pairs of vehicles that may be involved in criminal activity. CBP agents also suggest that criminal vehicles may cross at certain times or ports to try and evade inspection. We propose to modify the MI formulation to include these heuristics by using law enforcement data from border-area jurisdictions. Statistical tests and selected cases judged by domain experts show that modified MI performs significantly better than classical MI in identifying potentially criminal vehicles.