AIExplainer
Machine Learning Deep Learning Intermediate 2 min read

What is shrinkage?

A statistical phenomenon where the predicted values of a model are more extreme than the actual values

Shrinkage occurs when a model is overfitting, meaning it is too closely fit to the training data, resulting in predictions that are too extreme or exaggerated

Imagine a person who always thinks it's going to be either extremely hot or cold outside, when in reality the temperature is usually mild. This person's predictions are like a model with shrinkage, exaggerating the actual values

A company uses a model to predict employee salaries based on their performance. However, the model is overfitting and predicts salaries that are too high or too low, resulting in inaccurate predictions. To address this, the company uses shrinkage to adjust the model's predictions and make them more realistic

Shrinkage is often addressed in statistical modeling by using techniques such as regularization or cross-validation to prevent overfitting and improve the model's predictive accuracy

Shrinkage is not the same as overfitting, although the two are related. Shrinkage is a specific phenomenon that occurs when a model is overfitting, but not all overfitting results in shrinkage

The concept of shrinkage was first introduced by statistician Charles Stein in the 1950s, as a way to address the problem of overfitting in statistical models

regression to the mean overfitting correction

Three products for different needs — explore what’s relevant to you.