CISC483 Lecture Notes - Lecture 23: Nonlinear Regression, Linear Regression, Data Mining

• School: University of Delaware
• Department: Computer and Information Sciences
• Course: CISC483
• Professor: Carberry Sandra

Document Summary

Regression and model trees i. linear models often do not fit the data well b. Nonlinear regression even more complicated than linear regression. Why not combine the advantages of trees with regression? i. Continue splitting until reach a stopping criteria a. b. standard deviation for every node is <5% of the standard deviation of the original training set each node contains only a few instances (usually around 4) If a regression tree, then compute the average value for the training instances that reach each leaf node. If a model tree, then compute a linear regression formula for the training instances that reach each leaf node. When computing sdr, compute it only for the instances that have this attribute value and then multiply the total by a. b. m = # of these instances that have attribute values. T = # of instances that reach this node.