What is RealValue?

RealValue is a machine learning project for predicting home prices in Toronto. Using TensorFlow convolutional neural networks in conjunction with a dense network component, owners can take a couple of pictures of their home, enter a few simple details, and receive an accurate price range of what their home is worth. This ease of use allows homeowners to be confident about their residential decisions and be more informed about the real estate market than ever before.

By only providing pictures of the house and some primary house numbers such as number of bedrooms, bathrooms and square footage, the goal was to create an algorithm that can predict a house’s price in the range of 20% error. For example, for a house truly worth 500,000, a predictive range of around 400k-600k is quite informative and marks the point of success for this project.
We created a network consisting of two branches: a CNN for input images, and a dense network for input statistical data. The images of the house were fed into the CNN, while numerical data for that house was fed into the dense network. Then, results from both branches were concatenated and passed through one fully-connected layer. Finally, a single value is outputted, the price of the house.