Machine learning model for the classification of municipalities by illicit crops in Colombia from 2010 to 2020

Introduction— The United Nations Office on Drugs and Crime (UNODC) classifies Colombia as one of the countries where drug trafficking and crime threaten the security, peace and development opportunities of its citizens. Objective— This article presents the application of the unsupervised K-means classification algorithm to categorize municipalities with coca cultivation presence in Colombia. Methodology- The CRISP-DM methodology was used for data mining, and the PCA (Principal Component Analysis) algorithm was used for the correlation of variables. Results— Multiple sources of information were used, such as: the number of hectares of coca per municipality, seizures, laboratories destroyed, manual eradication and fumigation, monitored by national institutions, in order to make crosses with socioeconomic and performance variables of the municipalities with coca crops in the period from 2010 to 2020. Based on the classification, the scenarios of each category were analyzed to find scenarios that allow elucidating the dynamics of the territories suffering from this scourge. Conclusions— It was found that the behavior of coca-producing municipalities responds mainly to 4 groups. It was also found that the municipality of Tumaco in Nariño does not fit into any category since it exceeds the production with respect to the other municipalities.