- Timeline
- Sep 2026 – present
- Team
- IDX Exchange data science interns
- My role
- Data preparation, modeling, and evaluation
- Skills
- Python, CRMLS data, Regression, Feature engineering
The goal
Build a model that estimates the sale price of California homes from CRMLS listing data, and evaluate it on sales it has never seen.
Highlights
- 01
Tested on 12,500+ unseen sales
All reported numbers come from a held-out set of more than 12,500 sales the model never trained on, so they reflect how it would behave on new listings.
0.911R² on held-out sales
- 02
Typically within about 8%
Median absolute percentage error is 8.46%: on a typical home, the estimate lands within roughly 8% of the actual sale price.
8.46%median APE
Looking forward
Next I want to understand where the model is least accurate, by region and price band, and make those errors visible rather than averaging them away.