Visual analytics for diagnosing spatial interpolation models

The development and widespread use of statistical learning models have brought the need for tools that help analysts diagnose, build, and refine those models. In this work, in particular, we focus on interpolation models, which spatially predict the value of a variable based on the values of its neighborhood. Investigating these results spatially or comparing them with other models at different levels of granularity is still a challenge for the analysts trying to understand and refine their models. To deal with that, we propose a visual analytics model-agnostic tool for facilitating the comparison and refinement of spatial models at different levels of granularity using interactive visualization techniques. The tool was built in collaboration with specialists who used it to diagnose and improve a spatial model for predicting residential real estate prices.