Lecture-12: Bias, Variance, Over fitting and Under fitting
-Bias, Variance, Overfitting and Underfitting are fundamental concepts for Machine Learning.
-Here I go through two examples those are linear regression and polynomial regression with different degrees that make these concepts very easy to understand.
-In order to derive the best fit line, we have to consider Bias and Variance.
-Bias: manse an error rate on the training data set.
-Variance: means error rate on testing data set.
-Overfitting: means that, a model has been trained (learned) on training data set well but the same model (trained model) is applied on test data set to get the more error.
In the case of Overfitting
low bias for train data set.
High variance for test data set.
-Underfitting means that a model hasn’t been trained (learned) properly on training data set getting more error (high bias) and apply the same model (trained model) on test data set to get the more error (high variance).
In the case of underfitting
High bias for train data set.
High variance for test data set.
-A Good Model should be select based on low bias & low variance.
i.e., Low bias for train data set
Low variance for test data set.
-When we create a model such that the error rate we get with respect to train data set must be similar to test dataset. Then only we can say that our model is genialized well means low bias and low variance.