Prediction model:

Predicting and identifying patterns of quarrels in the city of Bogota is a fundamental task for the design of security policies and the efficient use of the city's available resources. In Bogota, brawls are the main cause of crimes such as homicide and personal injury with 70% in 2018 according to the security secretariat. Recently, methods based on hot spots, spatiotemporal point processes and multivariate regression have been proposed to predict crime events and model their dynamics using historical records. Usually these methods are adapted to the conditions and dynamics of a given city so their use for different environments is not immediate. The vast majority of these methods do not take into account the experience and knowledge of the experts of the institution in charge of making decisions, allocating surveillance resources and developing social intervention policies. In this presentation we show a new approach that combines the analysis of multiple time series and the knowledge of the expert user to make predictions of brawls in the city of Bogota. This approach has the advantage of being more adaptable to the changing conditions of brawls experienced by the user in charge on a daily basis. Additionally, it is important to consider that the time series are easily interpretable, which allows the user to engage with the prediction system to find extra knowledge in the exploration of the data and prediction results in an interactive way. The advantages of this design allow the user to make more informed decisions. On the other hand, the weaknesses offered by the time series that affect its performance such as: parameter settings, selection of appropriate time windows for forecasting and training, and the emergence of outliers are mitigated by the intervention, experience and knowledge of the expert user through guidance and orientation techniques in the forecasting system. This work shows that the prediction results present low prediction errors and allow finding quarrel dynamics at different time scales.