Graph Restrictions for Signal Processing of Homicides Data

Relative to other types of crime, homicides are statistically more challenging due to its sparsity and low frequency. To overcome this problem and based on previous works, we propose a methodology that combines the signal processing capabilities of the Graph Laplacian of Gaussian (GLoG) filter and the inclusion of related types of crimes with kernel warping. We compare our model with the results obtained in those previous works and with benchmark models from other papers. We found that the combination of both methodologies results in an improvement of the model and fares well with the benchmark models.