Literature Review Spatial Prediction

This paper identifies the need for politicians, planners and social scientist to be provided the tools to clarify and manipulate spatial distributions to predict future developments. Bayesian statistics offers a way to estimate values of a variable at locations that are not sampled. The paper tries to address a case where Tobler’s law is not applicable. They are using Marcov Chain Monte Carlo analysis to develop better model based on probability distributions. and This paper considers two versions of the Bayesian variant of a flexible semi-parametric model (a mixture of experts). This consists of a gate model which divides the problem into regions of local models called expert predicts. The expert predict is responsible for providing the output. The two methods involve either providing an input variable to a single expert or to multiple experts. These two Bayesian inference models were tested using real geographic data. The results show that both Bayesian models were able to capture stochastic relation in spatial data. The multivariable approach was able to identified nonlinearities not evident from the single dimensional approach. There were no comparisons of performance or memory usage offered by this paper. The experiments didn’t involve high dimensional dataset which could be a concern. They tried to long term predict illness based on Manchester, UK census data.