skip to content
Insights Blog

Three Cases for Machine Learning

Insights - Machine Learning - Gannett Fleming.
Author: Sam Brazil, Application Developer, GeoDecisions

With the ever-increasing creation of data, businesses are discovering innovative approaches to big data analysis to gain further insights to improve operations and better serve customers. Machine learning, a branch of artificial intelligence, is the scientific study of algorithms enabling computers to make decisions with minimal human intervention. Based on analytical model building, data scientists use machine learning to process and understand insights faster by recognizing underlying patterns in the data that are not always apparent to the human eye. The data science team at GeoDecisions—the geospatial division of Gannett Fleming—has been exploring machine learning through the frame of location. Some of our recent data science projects include neural network analysis of crash data, real estate classification, and optical character recognition for improved data entry processing.

Neural Network Analysis

One of our first experiments with neural network analysis involved using existing GIS crash data from a state department of transportation (DOT) to predict when and where a crash would be more likely to occur. After training the neural network model with the available data, the team recognized problem zones and identified opportunities to improve safety. Recommendations included where to deploy additional resources during inclement weather events and where to reinforce areas with higher probabilities of DUI crashes. Understanding patterns of how all the historic crashes connect, logistics coordinators can expedite incident response time and provide decision-makers with valuable lifesaving data to mitigate accidents.

Real Estate Classification

Another application the data science team designed was a pricing tool for a real estate agency to use when consulting with sellers to determine their listing price. The inputs for the model included both recent and historical house sales data from the multiple listing service (MLS) database. GeoDecisions geo-enriched the MLS data with additional socioeconomic and other GIS data layers. Using a random forest classification algorithm, the team efficiently modeled thousands of variables to determine whether a particular list price would likely result in a sale within one week.

What’s Next?

The intersection of geospatial data and artificial intelligence is unlocking the power of our data in many unexpected ways. Predictive models are allowing us to optimize resource allocation and maximize our ability flexibility and our ability to plan. Every day computer vision and image classification algorithms are surpassing the accuracy of the human eye. Deep learning and big data allow us to extract information and features from satellite imagery instantaneously, a task that was once manual and tedious. Though on their own AI and GIS are well-established fields, the combination of the two areas is still in its infancy. As the industry embraces this data-driven confluence, the GeoDecisions data science team is ready for the new challenges and opportunities to innovate.

Similar Blogs: Business and Technology Solutions, Business and Technology Solutions: Business Improvement and Innovation, Company, Corporate, Corporate/Industrial, Information Technology: Geospatial, Operations, ITS & Traffic: Incident & Emergency Management (IM/EM), Information Technology, Information Technology: Information Technology Services, Innovation, Transportation: Operations, ITS & Traffic, State & Local Government, Operations, ITS & Traffic: Traffic & Safety Engineering, Transportation