Machine learning-Bias and Variance

Bias: Error in the Training set

Variance: Error in the testing set

Regression problem

1.Underfitting

Model does not work well in the Training dataset and Test dataset. High bias and High Variance

Purple dots indicate Training set, Blue dots indicate Testing set.

2.Correct model

Model works well in Training and Testing set. Low Bias and Low variance

Purple dots and Blue dots coinciding each other indicate Training and Testing set

3.Overfitting

Model works well in the Training set, but does not work well with the Testing set. Low Bias and High Variance

Purple dots training set and blue dots indicate testing set

Classification problem:

1.Underfitting:

Training error:25%

Testing error:26%

Model does not work well with the training set and testing set. High bias and High Variance

2.Correct Model:

Training error<10%

Testing error<10%

Model works well with training and testing set. Low Bias and Low variance

3.Overfitting

Training error:1%

Testing error:20%

Model works well with training set but does not work well with test set. Low Bias and High Variance