Prediction based on time-series of aggressive behaviors. A case study Bogotá, Colombia

Predicting and understanding crime patterns are highly relevant tasks in citizen security planning. In partic-ular, understanding the Spatio-temporal dynamics related to aggressive behaviors is essential to design successful actions to mitigate more severe crimes such as homicide and personal injury. This work proposes a novel approach supported by historical data to predict criminal incidents related to ag-gressive behaviors. The model predicts occurrences of these behaviors based on a periodicity analysis on multiple time series of different scales. Our results suggest that this approach may predict events of aggressive behavior with high levels of precision at varying scales in space and time. Specifically, a relative error of 9.8% on average was obtained, outperforming related approaches.