Linear Regression

Abstract

Regression aims at predicting a future value, so the outcome we are trying to predict is a number, not a class. Many problems can be reduced to predicting a number; for example, predicting the median house value, or predicting the number of people who will be infected by a virus, or predicting the rate of readmission to a hospital in a certain season, etc. In all these examples, our outcome is a number, so regression can be used to predict the outcome. In other words, we can use regression to build a model that can predict (with a certain likelihood of success) the outcome based on the existing features (i.e., dataset attributes). The situation resembles estimating a function f that takes many variables as an input and computes a number that estimates (with a certain likelihood of success) what the future outcome will be. The statement “with a certain likelihood of success” refers to the probability that the model (i.e., function) is correct; that model’s probability of success can be computed when we build the model.