Evaluate your Regression model like a Pro! Understand these Evaluation Metrics and make faster decisions on your models.

In order to understand how your Regression model performs we evaluate the model on both the training and test data. Evaluate here means using a metric to quantify the Regression model performance using just a single number.

Underfit Model

The fit of the model can be understood well by comparing these evaluation metrics for both train and test data.

Consider initial data= Information + NoiseHere in the above case the resulting underfit model will have less of information and no noise.

Overfit Model

Consider initial data= Information + Noise Here in the above case the resulting overfit model will have everything information as well as noise.

Hence the aim is the find the model which best fits the data ie only the entire information with no noise.

Evaluating Metrics

R -Squared

1. R-squared is the measure of percentage of variance in the target variable explained by the model.

2. First metric to look at for linear regression to see the model performance.

3. If R-squared is higher, its considered better.

Adjusted R-squared

1. Adjusted R-Squared is similar to R-Squared but penalizes for the addition of many variables.
2. This metric is used when there are too many variables. When more variables are added the R-Squared value always increases but the Adjusted R-Squared does not increase.
3. If Adjusted R-squared is higher, its considered better.

Mean Absolute Error

1. MAE has the same unit as the target variable and it is the simplest metric to check how accurate the prediction is.
2. If there are outliers the error does not increase too much. Hence it is not sensitive to outliers.
3. Optimization of this metric is difficult only from a math point of view.
4. If MAE is lower, its considered better.

Root Mean Square Error

1. RMSE also has the same unit as the target variable and it is also used to fing how accurate the predictions of the model are.
2. If there are outliers the errors are magnified because of the square function in the formula. Hence it is sensitive to outliers.
3. If RMSE is lower, its considered better.

Thank for reading my article.

Designs for equations and tables— I tried them with the help of Canva

Source : https://people.duke.edu/~rnau/compare.htm