We developed a neural network that is trained to predict the prices of houses.
Reali is a US-based startup that disrupts the traditional real estate model by introducing a flat fee commission for buying or selling a house. Reali uses technology, which streamlines the buying and selling process and allows customers to get professional support from their real estate expert team. Price predictor is one of the core features of the Reali platform that supports the bidding process with more AI-powered decision making.
How to ensure accuracy of prediction results based on data about houses as well as historical data about transactions?
Our initial cooperation began with a research project that we allocated one month to achieve a 95% prediction precision over the sample dataset. The initial dataset was split into training and test subsets to test the accuracy of the predictions.
We used linear regression to understand how listing prices, area, year of construction, condition, and the number of rooms impacted the selling price of the house. Another useful insight that we uncovered was that some data entries from the initial dataset contained errors that resulted in low performance within the model. To solve this issue, we created a script that filtered incorrect input data.
As an outcome of this project, we have trained a set of models that were very close to the initial target that made it viable from the business perspective to invest further in this direction. In addition to that, we learned valuable business insights as well as challenges that should be addressed in the production version of the price predictor.
With over two years of cooperation between Reali and Temy, our joint team has created multiple neural networks that support the expert team with pricing offer preparation as well as listing price recommendations for the sellers. Tech stack: Python, Jupiter, Skala, Akka, Angular, Typescript