Metrics: RMSE, MAE and R²

Root mean squared error: $\text{RMSE}=\sqrt{\frac{1}{n}{\sum }_i(y_i-{\hat{y}}_i{)}^2}$. In the same units as the target, so it reads naturally ('off by about 12,000 on average'), but large errors dominate it.

Mean absolute error: $\text{MAE}=\frac{1}{n}{\sum }_i|y_i-{\hat{y}}_i|$. Also in target units, and more robust to outliers: an error of 100 counts 10 times an error of 10, not 100 times. RMSE ≥ MAE always; a big gap means a few large errors.

Coefficient of determination: $R^2=1-\frac{{\sum }_i(y_i-{\hat{y}}_i{)}^2}{{\sum }_i(y_i-\bar{y}{)}^2}$ — the fraction of the target's variance the model explains. $R^2=1$ is perfect; $R^2=0$ is no better than always predicting the mean; negative values are worse than the mean.