DSLab WNE UW

Aim of the project

The main goal of this project is effective software for spatial regression models in R enabling them to be estimated on large data sets.

Spatial regression models are used on data on areas (e.g. administrative territorial units – poviats or communes) or points in space (e.g. location of people, companies or other facilities). Objects located close to each other in space usually have an influence on each other, which is not automatically included in traditional statistical models or machine learning algorithms. Spatial regression models take into account the relationship between phenomena in space – e.g. impact on the value of a dependent variable in the model, the value of the same variable for neighboring objects, or the values of other features of neighboring objects.

The planned result of the project will be the R package allowing for the effective use of spatial regression models on large data sets.

Relations in space are taken into account using a square distance matrix that describes the relationship between objects and has the dimension nxn, where n is the number of observations in the model. Although the distance matrix contains many zeros (it is a rare matrix), optimization of spatial regression models for large data sets is a big challenge.
The planned result of the project will be the R package allowing for the effective use of spatial regression models on large data sets.