Housing prices are of great importance to many people: house owners, buyers, investors, and bankers, to name just a few. Open data by Statistics Finland provides average prices for areas with enough transactions. In a sparsely populated country like Finland, it means the prices are publicly known in only a small part of the country. Reaktor AI has built a service, kannattaakokauppa.fi, that uses open demographic data and state-of-the-art probabilistic machine learning to fill the gaps.

Publicly available housing price data covers just a fraction of the postal areas in Finland. And even in areas for which prices are reported, there might be gaps or the number of transactions might be so small that signal to noise ratio is weak. Simultaneously housing prices are sensitive to small changes in the environment and should still be known with a fairly high precision to be of use.

A probabilistic machine learning solution was built to estimate the actual price levels and changes in them on all the Finnish inhabited postal areas by utilising both demographic information and spatial hierarchy to transfer information from known prices. The solution pays great attention to taking uncertainties properly into account and visualising the information in an easily digestible format. This allows users to get the estimates themselves and have an idea of how confident the model is of an estimate.