ML Exam 2 - Module 10 : Nonlinear Regression

Regression Tree

Used to predict numeric values instead of class labels. The feature space is partitioned into regions that are used to predict a numeric value for each region.

Error

Used to build the tree (SSE) and evaluate the tree (MSE, RMSE, MAE)

Multiple Predictors

For each feature/threshold, compute the SSE after the splut and choose the split with the most minimal error. Used to predicted the error (duh)

KNN Regression

A k-nearest neighbours algorithm used to predict numbers rather than class labels. The k nearest training points are picked via euclidean distance and then subsequently averaged out.

Support Vector Regression (SVR)

A regression version of SVM that tries to fit a simple, kernel-based function that is as flat as possible while simultaneously keeping most prediction errors within a small tolerance. Basically trying to keep prediction errors with the smallest margin possible while penalising only larger errors.