Spatial feature engineering refers to using geographic information to “construct” new data or develop additional information from geographic information.

For example, given a dataset of metro stations, generate metrics for the number of restaurants and theaters within several distances. Or, given a neighborhood, generate the average house price surrounding a specific house or location. The results then are usually either numerical or categorical variables. These are also called spatially explicit independent or exogenous variables in the spatial modeling context. The purpose of spatial feature engineering is to generate those new features from spatial data, which can be treated as extra independent variables and be directly fed into general machine learning algorithms (without modification of the algorithms) for analysis and predictions.

Machine learning based on spatial feature engineering reflects that those processes are not the same everywhere geographically. This is one type of spatial machine learning. However, depending on the application cases, this may not fully consider more intrinsic spatial relationships or neighborhood effects. Thus an application may require some specialized machine learning algorithms provided by Spatial AI.

You can leverage Oracle Spatial database functionalities to engineer new features based on spatial data. In addition, the following three new feature engineering methods are supported - Spatial Lag Transformer, Categorical Lag Transformer, and Spatial Coordinates Transformer.