Imagine: a market value assessment report on any property in Québec, automatically generated in a few moments!
JLR, a land solutions company, sought CRIM’s help to optimize the operation of its property value analysis software, called Évia. This platform allows real estate brokers or individuals to obtain a market value assessment report on the property of their choice, generated automatically in a few moments from a Québec address.
How the software works
This application was originally designed using a statistical model that analyzes the data and combines it to provide a range of values as well as a median value for similar properties. JLR now wishes to refine its model and streamline its workings.
CRIM has been working on a new operating structure for the software. Instead of the statistical model, this structure is based on machine learning and neural networks.
Computer vision and satellite imagery
JLR currently holds over a million photos of building facades in its database and is in the process of generating a complete capture of the province through its Jakarto project. CRIM developed an algorithm that can use these photos to detect certain physical characteristics of the property: exterior cladding, size of windows, presence of a garage, etc. The algorithm can be used to determine the property’s physical characteristics. This information will be used to automatically complete the content of the report provided by the Évia platform.
Data science and predictive models
CRIM’s data scientists then integrated the results of image work and combined them with the data already available in the real estate database built by JLR over 30 years ago, in order to develop market value prediction models that are much more accurate than the current approach. The new model is designed to work even when data on the building of interest is missing. It can also flag transactions that deviate from the current trend for a sector.
The collaboration with CRIM has enabled JLR to replace its current statistical approaches with machine learning processes to obtain more accurate results on the value of each property, while increasing the robustness of the system and creating a higher value-added tool for its clients.