Experimenting with parameters that support early stopping
Overfitting models is one of the common issues often faced when trying to solve an ML problem. Overfitting is said to have occurred when the ML model tries to adapt to your training set too much, so much so that it is only able to make predictions on values that it has seen before in the training set and is unable to make a generalized prediction on unseen data.
Overfitting occurs due to a variety of reasons, one of them being that the model learns so much from the dataset that it even incorporates and learns the noise in the dataset. This learning negatively impacts predictions on new data that may not have that noise. So, how do we tackle this issue and prevent the model from overfitting? Stop the model early before it learns the noise.
In the following sub-sections, we shall understand what early stopping is and how it is done. Then, we will learn how the early stopping parameters offered by H2O AutoML work.
