Header image

How OhMyHome Used Machine Learning for Real Estate Valuation

September 8, 2021 case-study artificial-intelligence machine-learning property-valuation gcp real-estate

We helped OhMyHome, Singapore’s leading property transaction ecosystem, by building them:

  • Custom Machine Learning Models for the Singapore and Malaysia housing markets that predict property value with as little as 4% deviation vs actual prices for Singapore and 10% deviation for Malaysia.
  • Automated ML Training Pipeline so the model runs on fresh data every week
  • API Endpoint Integration with client’s native platform
On average only 4% difference against historical sales prices for Singapore

By reducing time spent on time-consuming analysis, the OhMyHome team can focus on what matters most: simplifying and improving the user experience of their clients.


OhMyHome (OMH) offers a one-stop solution for buyers, sellers, landlords, and tenants in Singapore and Malaysia, and has facilitated thousands of house transactions in the less than five years since they were founded.

To do this, they need a key piece of information: the fair value of the property. The traditional way of determining this is by looking at the property’s features (such as size, year built, or available amenities), compiling the neighborhood’s statistics (such as population or median household income), or monitoring economic indicators of the country’s housing market.

But collecting, compiling, and analyzing all this data is tedious and time-consuming -- and by the time you’ve finished it, your data could already be out-of-date.


With our expertise in geospatial analytics and artificial intelligence, Thinking Machines built a custom machine learning model that can estimate the price of any property in Singapore or Malaysia.

Harnessing a Wealth of Open-Source Geospatial Data

We trained the model to predict price by taking OhMyHome’s proprietary dataset of historical transactions and supplementing it with Thinking Machines’ extensive database of geospatial points of interest. This captures how the price of a property is affected by its surroundings. For example, our model can factor in the distance to the nearest major road or the density of restaurants in the surrounding.

Determining Price Drivers with Machine Learning

We then worked with OhMyHome to create a series of relevant features and input variables based on the domain knowledge and experience of their agents. The models were tailored for each housing market, Singapore and Malaysia. Our analysis also showed that landed properties (e.g. house and lot) had to be separated from non-landed properties (e.g. condominium units).

Although the models were localized, there were some consistent trends. As expected, units that were bigger, newer, or at a higher floor tended to have higher prices. However, location features such as the postal district, the distance to the nearest point of interest, and the density of nearby points of interest were equally important.

We evaluated the model by comparing the generated prices from their past transactions in 2019. Purchase Price Deviation is the percentage difference of the prediction vs the actual selling price. The model output had on average only 4% difference against actual sales prices for Singapore and 10% for Malaysia, when comparing against historical property transactions; meaning the model is very accurate in predicting property prices.*

*test data is for the Singapore HDB (Public) Market and the Malaysia Condo market

Putting It Into Production

A one-time analysis is not enough. After all, as new schools, restaurants, hospitals, or other points of interest are built, the value of the surrounding property could fluctuate. To make sure that our tool was robust and usable, we created a pipeline that would ingest fresh data every week and retrain the model for the most up-to-date estimates.

The entire pipeline encompassing data storage, data cleaning and transformation, machine learning, and the API were built using Google Cloud Platform and connected to OMH’s existing platform. This ensures that our model is integrated into their workflow and can be connected to other operational APIs within OMH’s infrastructure.

Data Preparation
API Integration

The finished Real Estate Valuation Tool allows OMH to input a property's details and immediately returns the estimated price, along with the valuation breakdown and a confidence rating of Low, Medium, or High.

This business solution gives them the data they need at an unprecedented speed and scale, freeing up their time for higher-value tasks and giving them an edge over their competitors.


How can our team help you get insights from data? Leave us a note on social media or email us directly at [email protected].


MAPPED: Danger Zones in Metro Manila's Roads

Which roads and intersections are the most accident prone for motorcycles? Where are pedestrians most vulnerable to be injured? Which routes are dangerous for both?

The Most Accident-prone Streets and Cities in Metro Manila (2014)

Part of our continuing series on road safety. Which roads and cities reported the highest number of persons injured or killed in traffic incidents in 2014?

Accelerated Customer Data Cleanup and Network Migration using AI for Globe Telecom

We built a custom Broadband Migration Tool that allowed Globe to automatically process and improve the quality of addresses at scale, resulting in a 6x increase in customer locatability and network matching.