HomeValuePro uses a gradient boosting regression model to predict apartment sale prices for agents in Chicago. The model is used to set listing guidance, and leadership is debating whether to optimize for RMSE or MAE after recent complaints about a small number of very bad predictions.
| Metric | Validation Set | Notes |
|---|---|---|
| MAE | $18,400 | Average absolute error across all listings |
| RMSE | $34,900 | Larger errors receive higher penalty |
| Median Absolute Error | $11,200 | Typical listing error is lower than MAE |
| Mean Prediction Bias | +$2,100 | Slight average overprediction |
| P90 Absolute Error | $52,000 | 10% of listings miss by more than this |
| Listings evaluated | 12,000 | Out-of-time validation sample |
Most predictions are reasonably close, but a small subset of luxury and newly renovated properties have very large misses. Product managers want to know whether MAE or RMSE better reflects model quality for this use case, and what the gap between the two metrics implies about the error distribution.