"My study involved looking at closed cases like where runaways had retuned, those who had suicided and those who were the victims of foul play. I looked at the background of each person, circumstances, letters and even newspaper articles, detectives' summary reports and even autopsy reports, microfiche and homicide reports."

Foy said that once the data was collated, it was analyzed using a basic comparative statistic. Then the entire data was mined using the j48part algorithm on Weka to identify a hierarchy of the most important predictive characteristics of missing persons.

Currently, the School of Information Technology at CSU is developing a user interface for the software to run on the police mainframe.

Foy said the goal of the project is using the search power of the mainframe to analyze data for officers in the field.

"The idea is to port this information into a mainframe for data mining and eventually look at using fuzzy logic to search the missing person data. Software is currently being developed so if police were to enter the details of a missing person into a hand-held device it would send the information wirelessly to the mainframe and the learning system would return a likely reason for a person going missing as well as a percentage risk factor," Foy said.