IN THIS CHAPTER
Illustrating linear models for regression and classification
Preparing the right data for linear models
Limiting the influence of less useful and redundant variables
Selecting effective feature subsets
Learning from big data by stochastic gradient descent
You can’t complete an overview of basic machine learning algorithms without discovering the linear models family, a common and excellent type of starting algorithm to use when trying to make predictions from data. Linear models comprise a wide family of models derived from statistical science, although just two of them, linear regression and logistic regression, are frequently mentioned and used.
Statisticians, econometricians, and scientists from many disciplines have long used linear models to confirm their theories by means of data validation and to obtain practical predictions. You can find an impressive number of books and papers about this family of models stocked in libraries and universities. This mass of literature discusses many applications as well as the sophisticated tests and statistical measures devised to check and validate the applicability of linear models to many types of data problems in detail.
Machine learning adherents adopted linear models early. However, because learning from data is such a practical discipline, machine learning separates linear models from everything related to statistics and keeps only the mathematical formulations. ...