A Classification Tree Based on Discriminant Functions

The classification problem is an important topic in knowledge discovery and machine learning. Traditional classification tree methods and their improvements have been discussed widely. This work proposes a new approach to construct decision trees based on discriminant functions which are learned using genetic programming. A discriminant function is a mathematical function for classifying data into a specific class. To learn discriminant functions effectively and efficiently, a distance-based fitness function for genetic programming is designed. After the set of discriminant functions for all classes is generated, a classifier is created as a binary decision tree with the Z-value measure to resolve the problem of ambiguity among discriminant functions. Several popular datasets from the UCI Repository were selected to illustrate the effectiveness of the proposed classifiers by comparing with previous methods. The results show that the proposed classification tree demonstrates high accuracy on the selected datasets.