Spatial Feature Engineering
Sergio Joseph Rey, Daniel Arribas‐Bel, Levi John Wolf · 2023
At its core, spatial feature engineering is the process of developing additional information from raw data using geographic knowledge. This distilling of information can occur between datasets, where geography is used to link information in separate datasets together; or within datasets, where geography can be used to augment the information available for one sample by borrowing from nearby ones. This chapter is structured following that distinction: for cases where geography connects different datasets, we adopt the term “Map Matching”, often used in industry; while we use the mirroring concept of “Map Synthesis” describing the use of geographical structure to derive new features from a given dataset. Technically speaking, some of the methods we review are similar across these two cases, or even the same; however they can be applied in the context of “matching” or “synthesis”, and we consider those conceptually different, hence their inclusion in both sections. Throughout the chapter, we use the AirBnB nightly rental prices in San Diego, as well as auxiliary datasets such as elevation or Census demographics.