Regression refers to a set of classical statistical models where a target variable is predicted using a weighted sum of predictors.
The most simple type of regression is Linear Regression, in which the mean of a continuous outcome is directly modeled as a linear combination of one or more predictors. Additionally, regression methods can be used to predict non-normally distributed response variables, such as a binary outcome (logistic regression), count data (Poisson Regression), or a proportion.
While regression has generally been more of a statistical technique than a machine learning algorithm, it can be used in prediction problems, and it often performs very well on linear decision boundaries and is a go-to method in cases where interpretation is important.