Industry Insights: SAS

Government Product News (GPN): What is Machine Learning, and how is the Wake County, N.C., Revenue Department using it?

Jennifer Robinson (JR): The Wake County Revenue Department uses the machine learning capabilities of SAS Viya. Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention.

SAS built a cloud-based, machine learning model for Wake County that considers thousands of factors and real-time property sales to offer timely, objective, highly accurate market forecasts. To accomplish this, SAS runs, revises and re-runs hundreds of thousands of decision trees.  With every property sale, the model is refined to become more precise.

County appraisers first perform their own analysis and determine values for each property, then turn to SAS Viya for an objective second opinion. Properties are currently selling quickly in the county, and these sales, in turn, influence market values for all properties. SAS Viya performs an independent, data-driven, objective analysis for each property that can be compared with county appraisers’ analysis and assumptions to validate accuracy or identify divergence.

GPN: What type of data does this generate, and how is this data helping the county?

JR: The Wake County Revenue Department strives to be efficient and effective by investing in staff when needed, and in cost-effective technology whenever possible. Taxpayers benefit through quality, accurate services for the lowest cost of operation possible.

With SAS Viya, appraisers can quickly identify neighborhoods that may require additional review to determine what is causing divergence and make any necessary adjustments. The goal is to make sure the new market values they assess on Jan. 1, 2020, are as fair, accurate and equitable as possible.

The system can predict a property’s sale price based solely on recent sales data, with no additional input from the user. There are dozens of inputs that affect property value: square footage, township, neighborhood, number of bathrooms, exterior finish, etc. A typical modeling approach would require substantive knowledge of how these inputs interact, for example, how square footage rates vary by neighborhood, or how the value of an additional bathroom is greater for houses with few bathrooms than for larger houses with many bathrooms.  However, using SAS machine learning software, such knowledge and assumptions are not required – the computer learns as it processes each home sale.

There are a lot of subjective factors involved when appraising a property so it’s helpful that the county now has a tool that only performs objective analysis on the data, providing a check and balance.   An appraiser may have worked in an area of the county for years and tended to rely more on their experience rather than what the data may be saying.  Sometimes this may be fine, but other times it may create some inequities. This is where a second, objective source is very useful.

GPN: What advice can you give similar departments who might want to use this same type of solution?

JR: First, the more sales data you have, the better. It will help refine the model. Also, it is vital to keep a centralized database that tracks both characteristics at the time of sale and present day.  Consider how as an appraiser your knowledge and experience can be quantified as measurable factors.  For example, you may know that rental properties sell for less than owner-occupied homes.  How do you determine if a property is rental? For Wake County, we accomplished this by comparing the physical address to the mailing address on record.

GPN: Are there other applications where these types of systems can be beneficial?

JR: Government agencies such as public safety and utilities have a particular need for machine learning since they have multiple sources of data that can be mined for insights. Analyzing Internet of things sensor data, for example, identifies ways to increase efficiency and save money. Machine learning can also help detect fraud and minimize identity theft.

Jennifer Robinson is the director of local government solutions for analytics provider SAS. She has served on the town council of Cary, N.C. since 1999. She has a particular interst in cities using information technology to improve the lives of their citizens and focuses much of her efforts on fostering the use of analytics in creating smart cities.