A Forensic System for Identifying the Suspects of a Crime with No Solid Material Evidences

Criminal investigators usually seek to short-list the suspects of a crime under consideration. This task becomes difficult when there are no clear witnesses nor material evidences identified by traditional means. In such a case, the investigators will try to identify potential suspects from the information of a pool of habitual criminals stored in the database. They manually create successively smaller and more tightly defined groups using a tier-structure of categorization attributes (e.g., location of offenses, category of crimes, age of criminals, etc.). We propose in this paper a digital forensic system called SISC that can automatically short-list suspects by categorizing the attributes of habitual criminals stored in the database using decision tree, logistic regression, and chi-squared analysis techniques. First, SISC constructs a decision tree by ranking the categorization attributes. It then identifies the path p (i.e., branch) in the tree that contains the potential suspects. SISC uses chi-squared analysis to identify the path p after employing logistic regression to estimate the linear decision boundaries of the categorization attributes. Usually, the leaf node in p contains the short-listed suspects, who are likely to have committed the crime under consideration. We evaluated the quality of SISC experimentally using real-world data. Results showed good prediction precision.