Density Estimation Based on Geographic Information

Xiangchu FENG,Zhaowei OUYANG,Chenping ZHAO,Xiaohui LI
DOI: https://doi.org/10.14188/j.1671-8836.2017.05.008
2017-01-01
Abstract:Given discrete events data, a probability density map can be produced to model the relative probability of events occurrence.Common methods such as Kernel Density Estimation do not take geographical information into account.In application, these methods could result in the support of probability density appearing in the unrealistic geographical locations.This article proposes Modified Maximum Penalized Likelihood Estimation methods based on Total Variation.It can both ensure the smoothness of the density estimation and guarantee that the density estimation of discrete events do not appear in the invalid location.This article verifies the superiority of the new algorithm by comparing the new algorithm with the traditional through applying simulation discrete data.Then, this article applies this method on criminal density estimation of a city to verify the feasibility of solving actual issue and giving police a guidance on deployment.
What problem does this paper attempt to address?