Types of Temporal Data - Feature Engineering

Feature Engineering from Geo-Spatial Data

April 4, 2023

Geospatial data, combining geographic and spatial information, is becoming increasingly important in various industries, from transportation and logistics to urban planning and environmental monitoring. However, working with geospatial data can be challenging, as it often requires specialized feature engineering and analysis techniques.

In this blog, we will explore some of the key concepts and techniques involved in feature engineering for geospatial tables.

Geospatial Tables 

Geo-spatial tables contain geographic data that can be used to represent geographic features. These tables are rows and columns containing information about geographic locations and features like roads, rivers, buildings, and parks. Each row represents a point in space or a specific area. The columns may contain longitude or latitude coordinates and other unique characteristics like environmental conditions, population density, and land use. 

For example, geospatial data showing land use in a town or city may have columns for longitudes and latitudes alongside other columns with land type use (this may include information such as whether the land is used for industrial, commercial, or residential purposes), the shape and size of each piece of land, and the land value.

The greatest challenge many experts face with geospatial data is that it is complex and highly dimensional, with many features that are challenging to visualize and interpret. This is where feature engineering comes in. It helps users quickly reduce the dimensionality of geospatial data to get the information they need for their machine-learning algorithms. 

IDLatitudeLongitudeBuilding TypeOccupancy…..
140.712776-74.0059742023-03-28 10:15:30AD475..…
234.052235-118.2436832023-03-28 11:25:45AX393..…
337.774929-122.4194182023-03-28 12:35:58AC304..…

Examples of Geo-spatial tables

Land Use Features 

Land use features can be used to capture the spatial characteristics of the data, such as the types and patterns of land use, the density of different types of buildings, or the presence of specific landmarks or amenities. Land use features can be particularly useful for urban planning and environmental monitoring applications. They can help identify patterns of land use change, the distribution of environmental resources, or the impact of urban development on local ecosystems.

For example, we might calculate the proportion of each neighborhood that is devoted to different types of land use, such as residential, commercial, or industrial. We might also calculate the density of different types of buildings or structures, such as high-rise apartments or single-family homes.

Distance-Based Features 

Distance-based features are a set of features derived from geospatial data that use the distance between two points as a measure of similarity. They can also capture trends and patterns in a data set. For example, we might calculate the travel distance of a taxi based on “pick-up location” and “drop-off location” (e.g., the famous NYC taxi dataset). We might also calculate the distance from a retail store to the nearest train station, which often significantly affects the demand patterns of the retail store.

An example of a distance-based feature

Spatial Aggregation and Autocorrelation

Spatial Aggregation combines data from multiple locations to create a new, more comprehensive dataset. This is often done to reduce the amount of data that needs to be processed or to improve the accuracy of the data by increasing the sample size.

Autocorrelation is the degree of similarity between data points that are close together in space or time. This can be a positive or negative correlation and is often used to predict a data point’s future values based on its importance. Identifying hotspots or clusters of activities and determining the patterns of heterogeneity or dependence is also crucial. 

A good example is where an analyst calculates the spatial autocorrelation of rents or land values and then uses the information to create unique features capturing the degree of dispersion or clustering in various areas. Spatial correlation can also determine the relationships between population and crime rates in different locations.

An example of a spatial aggregation feature

Spatial Interaction

Spatial interaction is the extent to which different geographical locations are interdependent or connected. This information can model how goods or people move and identify connectivity or accessibility patterns. 

For example, the spatial interaction between various commercial centers or transportation hubs can capture the extent of connectivity or accessibility in multiple locations. The spatial interaction can also identify spatial heterogeneity or dependence patterns. This is particularly important in the logistics and transportation industry, where it is used for routing optimization and scheduling decisions. 

Grid Target Encoding

Grid target encoding combines a grid-based approach with target encoding. This is a popular technique used in machine learning to encode categorical variables. Target encoding involves replacing categories with the median of mean target values for the respective categories. The analysts usually divide spatial regions into grids of cells and then calculate different features or statistics for the cells. 

Note that in grid target encoding, the target encoding is applied to all categorical variables, and then the resulting encoding is used to calculate each grid cell’s feature. Analysts can calculate the median or mean target value of the cell observations. This results in a grid-based feature for each categorical variable, which can then act as the input for a machine-learning model.

Grid target encoding is essential when observations have a complex or non-linear relationship. It can help simplify the data and make interpretation easier.

An example of a grid target encoding feature


Feature engineering from geospatial data is a powerful tool that can be used in various industries to improve the performance of machine learning models. Experts use feature engineering for geospatial tables to extract helpful information from spatial data. Though using geospatial data is challenging, using the right techniques and tools can help you gain valuable insights to help you make informed decisions.

You are not alone in this. You can rely on dotData’s Feature Discovery Platform for geo-temporal or geospatial table analysis. This platform automatically extracts grid target encoding features, land use, spatial autocorrelation, and distance-based data – and that’s just the beginning. You can visit the platform to learn more and ask any questions you may have.

Share On

Yukitaka Kusumura

Yukitaka Kusumura

Yukitaka is the principal research engineer and a co-founder of dotData, where he leads the R&D of AI-powered feature engineering technology. He has over ten years of experience in research related to data science, including machine learning, natural language processing, and big data engineering. Prior to joining dotData, Yukitaka was a principal researcher at NEC Corporation. He led the invention of cutting-edge technologies related to automated feature engineering from various data sources and worked with clients as a data science practitioner. Yukitaka received his Ph.D. degree in Engineering from Osaka University.