AI in real estate 4: How Healthy Workers algorithms achieved state-of-the-art
We now know that heating, ventilation, and air conditioning (HVAC) are the largest energy consumers in commercial buildings. To reduce emissions, it’s important to address it, and balance is key. We can’t simply reduce everything to zero. Workspaces must also remain healthy and comfortable. Thanks to research on how to combine the best of both worlds, we created a state-of-the-art algorithm. Find out what that means.
![](https://cdn.prod.website-files.com/668f916f1023297c4582c0f4/66cf25bb4578595471479d5c_AI%20in%20real%20estate%20-%20chapter%204%20header.webp)
Developing a machine learning algorithm in collaboration with the VU
We constantly strive to find the balance between reducing emissions and creating healthy and comfortable buildings. Scientific research is part of that quest. By predicting the indoor temperature of buildings, we can optimize HVAC.
Last year, in collaboration with the Vrije Universiteit Amsterdam, we conducted research on optimizing our machine learning algorithms. The result? A state-of-the-art model.
How it all began
In January 2023, we hired Ruthu Rooparaghunath at Healthy Workers. This enthusiastic AI master's student joined our research on developing our machine learning algorithm.
Ruthu took the lead in the research but worked together with co-founder Jonatan Roose, machine learning engineer Enzo van Kessel, and her professor Victor de Boer.
After graduating, she remained employed from July to December to implement her research and turn her thesis into a publication.
An accurate and state-of-the-art machine learning model
Our product is state-of-the-art thanks to several factors.
- Impressive accuracy of our algorithm: 94% accuracy of the algorithm. Meaning, 94% of our predictions of the indoor climate are within a 0.5-degree deviation from the actual indoor climate. If we make predictions for 24 hours for one hundred days, you've collected 2,400 predictions. Of those, 94% are correct, with a maximum deviation of 0.5 degrees from reality.
- We use one model: We train the model on data of a specific building, but ultimately, our model works on any type of building. This is remarkable and much cheaper than writing a new model every time, which is currently the standard in research in this area.
- We use real data: Existing research mainly uses simulated data. Not with us, we use real data. Sure, it takes more time, but fortunately, we started collecting it long ago.
Ultimately, this allows you to build a much more robust and fairer model.
It’s official: a state-of-the-art model for a greener future
After much hard work, the time has finally come. The research has been published, and we proudly say we have achieved state-of-the-art. This wonderful achievement holds great promise for a greener future for buildings.
As a company, we remain committed to scientific research—and ambitious master's thesis students are more than welcome. This is how we want to promote sustainability and show that we are leading the way towards net zero by 2050.
–
In the AI in real estate blog series, we’ll take you along in the rapid developments of AI and machine learning in the coming period. We will explain how it is changing the real estate world forever, what our vision is, and tell you about our AI developments concerning autonomously managed buildings.
Sign up for our monthly newsletter
Be the first to know about industry news, features and releases.
Discover related articles
Learn more about smart building optimizations from the experts.