Criminal activity negatively affects people's quality of life and economic progress. Given the advance in economic research, which leverages machine learning to detect patterns and analyze trends in specific fields, these techniques are being used in various contexts, including crime prevention. The objective of this work is to study the spatial patterns of crimes through the implementation of machine learning techniques, to predict the probability of occurrence of various types of crimes at an annual level with spatial differences in Medellín, Colombia, based on historical data. and sociodemographic. To carry out this objective, the Ordinary Least Squares, Random Forest and Extreme Gradient Boosting models were used, which obtained acceptable levels of performance, given their high precision. A relevant result is that the socioeconomic variables related to the proportion of men, people between 16 and 30 years of age, proportion of unemployed people, people who belong to SISBEN, proportion of people with multidimensional poverty, proportion of people with quantitative deficit of housing and who are part of socioeconomic stratum 1 or 2, both at the neighborhood and grid level had a high predictive power. For the purpose of this research, this will be used in decision-making and the formulation of public policies aimed at reducing crime.
