Evaluating regression metrics
Metrics in classification experiments, such as precision and recall, were oriented around sorting data into the right categories. Regression metrics are different and focus on how close our predicted values are to the correct value.
The difference between our model’s prediction and the actual value that occurred is referred to as the model’s error.
Understanding our model’s error helps us know both how accurate it generally is and how sensitive it can be to outliers.
The error can be positive or negative based on whether the model guessed a number greater than or less than the actual value. This is why metrics for regression involve getting the absolute value of error or squaring that error for each row so positive or negative values don’t matter.
Let’s cover the three simplest regression metrics first, as these directly measure error:
- Mean Absolute Error (MAE) is the simplest metric. If you add up...
