Proportunity: AI Helps First-Time Buyers

An authorized and regulated mortgage lender based in London is working with Intel to predict future house prices.

At a glance:

  • Proportunity is an innovative mortgage lender leveraging machine learning (ML) to make home ownership achievable for more Londoners. Proportunity uses ML to train models that can accurately forecast house prices region by region and identify those with growing value.

  • After testing various CPU- and GPU-based systems, the company now relies on Intel® Xeon® Scalable processors and Intel®-based Google Cloud instances for both its AI training and inference.

author-image

By

House prices in London are notoriously high—the average is almost double that of the UK as a whole—making it very difficult for first-time buyers to get onto the housing ladder. Proportunity has built a solution aimed at tackling this problem.

By performing predictive analysis on a deep web of different datasets, Proportunity can accurately identify homes and up-and-coming neighborhoods that are good investments—and then offer equity loans against the future price. With a Proportunity loan, borrowers can boost their down payment by up to 25 percent of the property price: a huge leg-up for first-time buyers.

Proportunity evaluates around 150 different factors, which it draws from a variety of different data sources: some built in-house, some publicly available, and some from third-party suppliers. By joining the Intel® AI Builders program, they’ve been able to accelerate their solution with access to the latest Intel® technology as well as engineering expertise.

“Thanks to Intel’s support, and access to the AI Builders program, we have been able to run our models faster and more efficiently. When using the Intel® DevCloud, our overall inferencing times improved noticeably, performing more than an order of magnitude faster, enabling first time buyers to get near real-time property insights on our platform.” —Sreekumar Balan, Research team lead, Proportunity

Spotting Diamonds in the Rough

Proportunity looks at data from two categories: property specific and area specific.

Property specific data characterizes individual houses. Prominent variables look at price history, surface and layout of the property, and its type (whether flat or house). Other data points factored in include the location of the property (postcode, longitude, latitude) and whether the property is a new build or not. For flats, number of floors in the building and the floor a property is on also has an influence.

Area specific data looks at less granular information, usually at the postcode level. The variables considered in Proportunity’s model include crime, unemployment, and disposable income. Demographic changes (e.g., change in generation X population) and the types of business (e.g., artisan coffee shops) in the area help models understand the gentrification potential of the area. Moreover, transport links—including future infrastructure projects such as the Bakerloo line extension—abundance of wellness facilities, and the presence of green spaces, provide valuable insights.

“Proximity to a tube station can increase a property’s value by five to ten percent.” —Stefan Boronea, cto, Proportunity.

Sourcing and Preparing Data for Use

Several datasets used by Proportunity are made publicly available by the government—property transactions data showing daily breakdowns of each transaction are published monthly, for example, as are energy performance certificates.

But Proportunity doesn’t just harvest open data: it does its own research. The company has a whole team dedicated to sourcing data that will impact house prices. This requires some creative thinking—for example, gathering information on scheduled completion dates of infrastructure projects from local boroughs.

The team is also currently working on integrating an innovative new data source into its calculations. By analyzing photographs of houses found online using computer vision technology, it’s possible to identify certain factors that impact value. These can include anything from the quality of paintwork to desirable features like bay windows.

Eliminating Bias and Ensuring Accuracy

Once the data has been sourced, it can take anywhere from a few days to a month to integrate into Proportunity’s predictive engine. During this period, it carries out a number of filtering techniques, including checking for outliers, as well as spatial and time smoothing (a process by which natural fluctuations in the data are accounted for).

Beyond the datasets, it’s also important to avoid introducing personal bias into the model—no assumptions are made by any one team member about what implications a data point might have for the value of a house. It’s intuitive to imagine that a high burglary rate would have a negative impact on house price, but the opposite can sometimes be true—criminals target more affluent areas. An open mind is essential.

When looking at future home values, Proportunity uses the data available in relation to other properties and areas. If crime in an area is rising, that will not automatically have a bad impact over its future value—crime across London could be rising at a higher overall rate, making the property more attractive.

To ensure the high levels of accuracy required to operate in the high-stakes property market, it is crucial that models are thoroughly tested before being used in the real world. That’s why Proportunity uses a technique known as “backtesting.” This involves training a model on historical data and observing whether the predictions it would have made in the past hold true.

Intel and Proportunity – Building a Winning Partnership Together

The housing market is volatile, which means that Proportunity uses multiple models at any point and is constantly prototyping new ones. Because of this, Proportunity’s ensemble of models prefer Intel® processors:

“With Proportunity being part of Google for Startups' Machine Learning program in London a few years back and with our ensemble of models favoring CPUs, it became a natural choice to build our infrastructure on a CPU-only Google Cloud Platform (GCP) cluster. Augmenting the value of already being a Google user, a GCP cluster allowed us to continue to develop on Intel Xeon Scalable processors, which are ideally suited for machine learning.” —Sreekumar Balan, Research team lead, Proportunity

Through participating in the Intel AI Builders program and using the Construction Zone Proportunity can rapidly spin up new models. It takes advantage of close contact with dedicated Intel engineers to help them optimize models, switch to optimized libraries, and integrate new AI technologies.

As the solution evolves further, Intel will continue to support the Proportunity team. You can find out more about the partnership by reading this blog on the Intel Builders website.