The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #92104

Title:An Optimal Decision Tree
Date: 1992
Abstract:The aim of this project was to develop an optimal procedure which would classify an unknown object X into the class Ci it belongs to. The classes are normally distributed and their statistics are known. The procedure has been chosen to be a decision tree, that will indicate the various features to be measured and in which order. It will also provide the user with some threshold values which, when compared with those obtained from the feature measurements, will give us a better idea of the true class of the object, and direct the search further. This decision tree will be optimal in the sense of giving least cost to the classification process. It will balance the cost of measuring the different features with the costs associated to the misclassification errors. Our program has been implemented in C, and will construct several decision trees, before comparing their costs and returning the optimal one as its result.

[Search These Pages] [DAI Home Page] [Comment]