A Predictive Model for the Missing People Problem

Abstract

The disappearance of people is a multidimensional phenomenon, in which several aspects must be considered. It affects people’s security perception and consumes police resources in its treatment. Therefore, does exists an emotional circumstance for the relatives of the missing person. At the same, the police departments must develop a search task, in most cases with much uncertainty. In this research, a predictive model to predict missing people’s status is presented. The information used to create the model come from the Colombian legal Medicine Institute, in a public dataset composed of 6202 cases and 11 variables. The output variable was the final disappearance status, with the categories Appears Dead, Appears Alive, and Still Disappeared. Three supervised machine-learning algorithms were trained and tested for the model creation, K-Nearest Neighbours, Decision Trees, and Random Forest. The study was divided into three phases, first considering all the output categories. In the second phase, generating a binary classification for the Appeared and Not appeared instance. Thirdly, models were built to predict the status of appeared persons, Appears Alive or Appears Dead. The K-NN algorithm outperforms the other models with an Area under the curve value of 94.8%.