Detection of criminal activity patterns through Machine Learning techniques

Abstract

San Andrés de Tumaco presents a public order situation that generates threats and extortions, being a  source  of   income for illegal armed groups (FIP - Ideas for Peace, 2022) and an instrument for

social and economic control. That is why this study proposes to carry out an application that contributes to the prediction of criminal activity in the municipality of Tumaco, as a tool to forecast the density of zoned criminal growth, as support to the actors in charge of security in the municipality to carry out mitigation plans and efficient use of the force. The public order situation that plagues some areas of the Nariño department, including the municipality of Tumaco; For its part, threats and extortions: a challenge to territorial peace, shows how "extortion networks configure mechanisms through which illegal or criminal armed groups appropriate the economic activities of the territories" (Defensoría Del Pueblo, 2022). The main objective of the project is to implement an experimental framework for comparing the commitment in quality metrics between quality metrics and computational cost in supervised and unsupervised Machine Learning algorithms to obtain a suboptimal model for predicting criminal activity for the municipality of Tumaco through the KDD methodology. The methodology was carried out using Knowledge Discovery in Databases (KDD) (Data, 2022). The results provide a crime database for all of Colombia for the years 2010 to 2019; transformation and cleaning scripts from which the training databases are obtained for both supervised and unsupervised techniques. It is concluded that the application of Kriging interpolation for the development of this research was the most favorable since the advanced geostatistical procedure generated an estimated surface from a set of criminal activity points dispersed with z values, thus obtaining an autocorrelation, that is, the statistical relationships between the measured points.