17
Classifiers: Decision Tree
§
§Well studied, widely used
§
§Implementation: “Computer and Robot Vision” (by R. Haralick and L. Shapiro)
–Top-down, greedy
–Decision rule: threshold a measurement component
–Threshold and component selection: entropy based purity measure
–Stopping criteria: maximum tree depth, minimum sample size
–
§
§
(1)
The decision tree algorithm used in our study was implemented from the description by Haralick and Shapiro in the book … It is similar to the widely used C4.5 algorithm: It uses a top-down greedy strategy, the decision rule at each internal node is thresholding one of the feature components; the selection of both the threshold and the feature component is based on an entropy based purity measure, and finally it uses the maximum tree depth and minimum sample size as stopping criteria.