Given all the available options and industry jargon surrounding data and analytics in real estate forecasting models, there can be confusion around machine learning and artificial intelligence. Machine learning is widely acknowledged as a form of artificial intelligence, but its overuse as a black-box approach in forecasting models can raise questions for real estate planners. Among the available modeling options, how do you know which is right for your brand?
Our team of data scientists conducted a study to test advanced machine learning modeling techniques and determine the effectiveness of each when applied to real estate forecasting for average size brands.
We examined some of the most commonly used modeling algorithms: statistical regression, nearest neighbor, decision trees, and random forest, to determine the advantages and disadvantages of each.
Using these algorithms in the study, our data scientists tested how a human-built, statistical regression model compared to strictly machine-learning built models for a restaurant brand with over 900 locations. We then compared the model results to the brand’s actual performance.
This evaluation helped identify differences in how each technique:
- Responded to validation
- Handled the addition of more variables
- Identified variables as key success drivers
- Determined the direction of a variable’s effect
The observations from this exercise help guide when to use machine learning on its own and when to include a statistician.
Read the complete study here.