Updated: Oct 24, 2022
In the quest to revolutionize real estate development, it pays to look into new technologies and experiment with them. The ability to recognize and adopt a successful trend early on can put a company at a significant advantage against its competitors, and give the entire industry a boost through spillovers. Now, Machine Learning (ML) isn’t exactly new (in fact, the first neural network dates back to 1958!), but it has been making a comeback in recent years due to the increased computing resources of the new century, including advancements in GPU technology. Today, almost every big name in the tech sector relies on ML to some extent. Giants like Alphabet and Meta have even developed their own in-house libraries, Tensorflow and Pytorch, which are now the gold standard for deep learning projects all over the world.
OMRT is currently taking part in the Hyperion Labs programme for AI and HPC start-ups
The benefits of ML are not limited to the tech sector. The idea at the core of ML development is that some decisions and measurements can be delegated to a machine, which can calculate the hidden factors behind each decision much faster than a human. This simple but powerful idea can be applied to almost every problem we face on a daily basis, and real estate development is no exception. In this article, I will tell you a little bit about one of the ways OMRT, like other companies, tries to make smarter decisions using ML.
Imagine you are a real estate developer. Think of the challenges involved in one of your
average projects. You have just bought a plot of land and you are about to fill it with buildings. Should you go for an apartment block, perhaps with a penthouse on top? Or maybe a row of houses? A building complex is a major commitment, so as a developer you want at least some assurance that the returns on your investment will exceed the costs, before you make a decision. You could, of course, spend months and months talking to consultants and experts, who would dust off their old textbooks, do an in-depth investigation into the market in your project’s location, and then come up with an itemized list of all the reasons why you should or shouldn’t build those buildings. Or you could use the ML approach.
The determinants of house prices have been investigated fairly extensively in the literature. It really boils down to the location, house type, and any special architectural features the building might have. Because architectural features are really hard to measure, it makes sense to consider them part of the random fluctuations around the average price. Information relative to the location and house type, however, is much easier to obtain. Starting from the postcode of a new building project, one can use APIs to extract all sorts of information about the area. In particular, city, region, and nearby amenities such as parks, malls, landmarks, parking spaces and transportation stations turn out to be pretty strong predictors of the house price.
Artificial Neural Networks aim to mimic the operation of a human brain
With these premises, you can see that it’s perfectly reasonable to think that we could let an algorithm estimate the potential house prices for a given postcode in our place. Data Scientists refer to such algorithms as “regression models”, and this is exactly what OMRT tried to do. We selected a list of approximately 1200 real estate projects widely distributed across the Netherlands, to capture as much variety as possible. For each postcode, we collected information on the city, province, house type, the number of parks, shopping centers, historical sites, parking spaces and transit hubs available via a 5-minute walk, and of course the asking price per square meter. With this data, we trained a Support Vector Regression model capable of predicting the price within a 10% margin of error.
Below you can find a small demonstration of the output of our model. Our model’s predictions for apartments in a neighborhood of the city of Arnhem are matched with the real borders of the corresponding postcode, and colored polygons are generated on a gradient ranging from green (very high price per square meter) to red (very low). The prices vary widely: the highest observed price was around 5000 euros per square meter (for a penthouse, of course), while the lowest was somewhere between 1000 and 2000 (for a hoekwoning, a typical Dutch building class). It is also interesting to observe some dynamics in the price. For example, all other things being equal, a unit increase in the number of parks within 10 minutes walking distance corresponds to an increase in price. This increase is massive (24%!) when you switch from 0 to 1 parks, then becomes gradually lower and eventually converges to an equilibrium level.
Sample output of the OMRT model
This is just one of countless applications of Machine Learning in the real estate development industry. The implications of these discoveries are significant: by delegating some of our tasks to machines, we can reduce the workload of developers, experts and consultants, who can instead fully dedicate themselves to more strategic decisions. This results in higher efficiency, lower costs, and better designs, which is what companies like OMRT stand for. We are witnessing a shift towards automation in the entire industry, and we hope that more and more businesses will join this movement to build a more sustainable future.