Real-Estate Valuation with Heterogeneous Geospatial Information

The proposed project investigates the extraction of geo-location information from online map services to augment the available attributes of real-estate properties, thereby enhancing model valuation of property sales price. Specifically, five distinct and concurrent objectives are proposed, each undertaken by a separate intern, to explore the feasibility of utilizing different data types retrieved from online map service, including satellite imaging, street-view photography, and traditional mapping of amenities. As these raw data are relatively complex, state-of-the-art machine learning technology, including convolutional network, graph neural network, attention mechanism, and objection detection models will be employed to extract meaningful geo-location information. Completion of this project will yield both commercial and technological benefits for the partner organization, as they control the distribution and access of the proprietary technology of automated property valuation.

Faculty Supervisor:

Di Niu

Student:

Partner:

Greentown Homes

Discipline:

Computer science

Sector:

Construction and infrastructure; Real estate and rental and leasing

University:

University of Alberta

Program: