Introducing DataRobot Location AI
With the incredible growth of machine learning and data science in the last decade, the realization that location is an important aspect of predictive modeling in many industries has become apparent. Businesses are quickly realizing that geospatial analysis and location intelligence are core to their predictive modeling workflows. These tasks range from simply visualizing features on interactive mapping tools, to working with native geospatial file formats, to advanced geospatial analysis, such as spatial querying, geoprocessing, and spatial lagging techniques.
Geospatial techniques like these have long been important for industries such as real estate, insurance, and retail, but demand is now starting to grow for incorporating them into emerging machine learning workflows.
Analysis of geospatial data has not only grown in prominence across many industries, but also across the workforce. Traditionally, the domain of Geographic Information Systems (GIS) Analysts, new business personas are now finding increased need to level up their geospatial analysis skills. The challenges these personas face are vast, and many of our customers have reported some recurring patterns. These include:
- Business analysts are stretching the geospatial capabilities of business intelligence tools to their limits, trying to conduct large-scale geospatial analysis and visualization with tools built for small-scale identity mapping and querying.
- Data scientists are finding that advanced spatial analysis techniques are only available in niche R and Python libraries and have limited exposure to the concepts familiar to a very small population of spatial data scientists.
- Data engineers are finding that geospatial analysis techniques available in spatial databases are difficult to connect to emerging machine learning pipelines.
- GIS professionals are running into limitations of including their traditional GIS-based tools and workflows into ML pipelines.
Enter DataRobot’s Location AI
Today, we are announcing the public beta release of Location AI for the DataRobot automated machine learning (AutoML) product. Location AI is a patent pending feature that enhances the DataRobot AutoML experience by incorporating a full range of geospatial modeling techniques. These include:
- Exploratory Geospatial Visualizations. Location AI provides an entirely new way of examining your data and exploring initial patterns using dynamic map visualizations.
- First Class Support for Spatial Data Formats. Through the same intuitive and easy to use automated machine learning UI that you work with today, you can now upload a range of native geospatial file formats, including ESRI Shapefiles, GeoJSON, PostGIS tables, and ESRI Geodatabase. DataRobot also automatically recognizes well-known geospatial data formats, such as latitude and longitude, lines and polygons. You can also select location features from your existing datasets.
- Spatially-Aware Feature Engineering and Models. Most AI models utilizing location take naive approaches to incorporating spatial data such as numeric or categorical representations. When DataRobot recognizes geospatial data types, it will perform specialized automatic feature engineering on that data to further enrich the dataset with additional spatial attributes such as the centroid, perimeter, area and minimum bounding rectangle (MBR) area, and so on.
We have also built explicit treatment of geospatial variables into our model blueprints via a novel Spatial Neighborhood Featurizer to fully realize the location of individual rows in your data and their spatial relationship to all other rows in the dataset.
- Unmatched Interpretability at the Local Level. Many businesses understand location is important to their models, but how important? Location AI gives users the tools to understand how their models are behaving at a local level. Feature importance and model accuracy can vary greatly across geographic locations. Our new “Accuracy Over Space” explainability visualization shows you exactly where, in terms of location, your model is accurate and where it is not.
Putting It All Together
One of the big differentiators for DataRobot’s automated machine learning product is its ability to handle multimodal datasets. By multimodal, we mean the ability to mix traditional tabular data containing numerics and categorical features with unstructured data like raw text, and images. Location AI further expands this concept and allows location data to be added to the mix. We make better decisions when we have perspective from more diverse types of information. AI is no different. With DataRobot, all of these diverse data types come together to help give your models the best possible accuracy from a broad range of features.
The example above shows a feature impact chart for a model we built to predict house prices in Utah. Here, all of these diverse data types including tabular data, text, and images come together with location to provide the best possible accuracy to the model.
Getting Started with Location AI
If you already have DataRobot automated machine learning, you can get started with Location AI right away. It is part of our new 6.1 Release and is available for all editions and all deployment options on-premises and in the cloud. No additional licenses are needed to use Location AI, you just need to enable the feature in your settings. Your DataRobot account team can help you with this if needed.
At DataRobot, we believe that the best results for building AI and machine learning applications come from collaborative team efforts. We encourage you to check out this important new capability, as well as some of the other exciting new features in our latest release. Simply connect to your data and start sharing content, creating projects and generating new predictions today.