QBUS3820 Lecture Notes - Lecture 6: Identity Matrix, Dividend Yield, Maximum A Posteriori Estimation
QBUS3820: Machine Learning and Data
Mining in Business
Lecture 6: Linear Model Selection and Regularisation
Semester 1, 2018
Discipline of Business Analytics, The University of Sydney Business School
find more resources at oneclass.com
find more resources at oneclass.com
Lecture 6: Linear Model Selection and Regularisation
1. Introduction
2. Subset selection methods
3. Regularisation methods
4. Discussion
2/49
find more resources at oneclass.com
find more resources at oneclass.com
Introduction
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Discipline of business analytics, the university of sydney business school. Lecture 6: linear model selection and regularisation: introduction, subset selection methods, regularisation methods, discussion. In this lecture we again focus the linear regression model. We move beyond ols to consider other estimation methods. The linear regression model is a special case based on a regression function of the form f (x) = 0 + 1x1 + 2x2 + . In the ols method, we select the coe cient values that minimise the residual sum of squares b ols = argmin. In the framework of the previous lecture, we can view ols as an empirical risk minimisation method. When the number of predictors, p, is large (e. g. comparable to n) ols has high variance and tends to over t. We can improve performance by setting some coe cients to zero or shrinking them. In other words, we will accept some bias in order reduce variance.