1 of 17

Metrics For Linear Regression

February 21, 2020

Phathutshedzo Ramakhanya

2 of 17

What is this all about?

  • Linear Regression Models are qualified by how good predicted values match up actual values, but how?
  • Statisticians developed error metrics to “judge” the quality of a model, which enables us to compare regressions of different parameters of interest.
  • Short and useful metrics to summarise the quality of data.

3 of 17

Proposed deliverables

Mean Squared Error

(MSE)

  • Measures average squared difference between estimated values and actual values

Root Mean Squared Error (RMSE)

  • Measures how large residuals are spread out

Mean Absolute Percentage Error

(MAPE)

  • Measure of prediction accuracy of a forecasting method in statistics

Mean Percentage Error

(MPE)

  • MAPE without the absolute value operation

Mean Absolute Error

(MAE)

  • Describes the typical magnitude of the residuals

4 of 17

Understanding the Metrics

5 of 17

Mean Absolute Error (MAE)

6 of 17

Mean Absolute Error (MAE)

7 of 17

Mean Squared Error (MSE)

8 of 17

Mean Squared Error (MSE)

9 of 17

Root Mean Squared Error (RMSE)

10 of 17

Mean Absolute Percentage Error (MAPE)

11 of 17

Mean Absolute Percentage Error (MAPE)

12 of 17

Mean Absolute Percentage Error (MAPE)

13 of 17

Mean Percentage Error (MPE)

14 of 17

Mean Percentage Error (MPE)

15 of 17

So, basically…

16 of 17

Questions?

Please be kind :)

Thank You!

(References on request)

17 of 17

Reference:

https://www.dataquest.io/blog/understanding-regression-error-metrics/