Model Evaluation Measures (Median Absolute Error, R^2 Score)

In this lesson we will look at different evaluation measures for Regression Models.

Median absolute error

Median is a Descriptive Statistics Measure and is robust to outliers in the dataset. The rate of error is calculated by taking the median of all absolute differences between the target and the prediction. Lower values means the model is more accurate.


If y^\hat{y} is the predicted target real valued output, then yy is the corresponding target real valued output and mm is the total number of instances. Then the median absolute error is calculated as follows:

MedAE(y^,y)=median(y^1y1,...,y^mym)MedAE(\hat{y}, y) = median(|\hat{y}^1 - y^1|,...,|\hat{y}^m - y^m|)


These code examples have been taken from the Scikit Learn Documentation.

Get hands-on with 1200+ tech skills courses.