Inferring time trends using machine learning

Time is an important factor in real estate appraisal. Analyzing the rate of change in property values becomes more challenging with heterogeneous types of properties and during uncertain times, such as the pandemic. At the Municipal Property Assessment Corporation (MPAC), in Ontario, Canada, we are utilizing the power of Machine Learning techniques to understand change trends in property values in Ontario. Through this approach, we can detect value change rates in a granular time dimension and by different geographical areas, property types and property attributes (size, quality of construction, age, etc.). Using an XGBoost (eXtreme Gradient Boosting) algorithm, we have trained a model on sales data for the properties sold over a period we want to detect the time adjusted factors (TAF). Machine Learning based time adjustments are then used in adjusting sales prices to the valuation date to build Ordinary Least Squares (OLS)regression models and generate ratio studies and time analysis reports.