Abstract:

This paper presents the concept of using ML methods to infer the reason of the people missing. The issue of determining the reason of disappearing of persons is extremely important and orientates police action as well as has huge impact on the effectiveness of the search. In consideration of the amount and diversity of information collected in the process of searching for missing persons, the police services should rely on templates that are both readable to humans and have a structured form for the purposes of computer support methods. The formal model, we propose for description of a missing person, ensures consistency and completeness of data. Based on historical cases, an aggregated model of disappearances was built. Six ML methods were selected and calibrated on the basis of historical cases, and then their quality was compared. Due to the relatively small set of real cases available in the system, 40 000 synthetic cases were generated. In order to identify the best method of classification, several selected methods were studied: Bayesian networks, naive Bayesian classifier, decision trees and random forests, and multilayer perceptron.