Crime risk assessment through Cox and self-exciting spatio-temporal point processes
Isabel Escudero,José M. Angulo,Jorge Mateu,Achmad Choiruddin
DOI: https://doi.org/10.1007/s00477-024-02857-2
IF: 3.821
2024-11-20
Stochastic Environmental Research and Risk Assessment
Abstract:Crime risk assessment needs tackling complex interrelationships between stochastic and deterministic components of spatio-temporal models. Criminal phenomena can be modeled using spatio-temporal point patterns of certain criminal data, and here we pay attention to the stochastic models of log-Gaussian Cox processes (LGCP) and self-exciting Hawkes processes (SEHP). We provide a comprehensive modeling strategy, combining both processes, noting that: (a) an LGCP facilitates the incorporation of first-order information through spatial and temporal deterministic components and second-order information through a stochastic component, and (b) a SEHP provides sufficient flexibility to incorporate various components in the background subprocess. To account for crime risk assessment, the deterministic components of the LGCP were estimated using a generalized linear model (GLM) for the temporal part, and a generalized additive model with B-splines for the highly nonlinear spatial covariates. In addition, the background rate components of the SEHP were estimated by a non-parametric stochastic reconstruction technique that includes a temporal periodicity, a separable spatial component, a long-term trend, and a semi-parametric method for the relaxation coefficients. MCMC-MALA and maximum likelihood were used for inference in both the LGCP and SEHP processes. We analyze crime events from the city of Riobamba (Ecuador), and with a complementary use of both stochastic point process models, we are able to assess the risk of crime, and provide reliable forecasts for weeks ahead.
environmental sciences,engineering, environmental,water resources, civil,statistics & probability