What technique is used to help prevent linear models from overfitting during training?

Regularization is a technique specifically designed to help prevent overfitting in linear models. It works by adding a penalty term to the loss function that the model is trying to minimize. This penalty term aims to constrain the model's complexity, effectively discouraging it from fitting the noise in the training data.

There are various forms of regularization, such as L1 (Lasso) and L2 (Ridge) regularization. L1 regularization can lead to sparse solutions by driving some coefficients to exactly zero, while L2 regularization tends to distribute the error among all features more evenly. By applying regularization, the model generalizes better to unseen data, which is crucial for avoiding overfitting.

Other techniques mentioned, like normalization, cross-validation, and subset selection, do play roles in machine learning but address different aspects of model training. Normalization typically focuses on scaling feature values to provide a more uniform data distribution, which does not directly prevent overfitting. Cross-validation is a method used to assess the performance of a model and ensure it generalizes well by training it on different subsets of data, but again, it does not inherently prevent overfitting during the training phase. Subset selection is about choosing specific features to include