Inducing Decision Trees based on a Cluster Quality Index

Decision trees are popular classifiers in data mining, artificial intelligence, and pattern recognition, because they are accurate and easy to comprehend. In this paper, we introduce a new procedure for inducing decision trees, to obtain trees that are more accurate, more compact, and more balanced. Each candidate split is evaluated using Rand Statistics, a quality index based on external measures, because it is considered by many authors as the best existing index. Our method was compared with other state-of-the-art methods and the results over 30 databases from the UCI Repository prove our claims. We also introduce a new equation to measure the balance of a binary tree.