Machine Learning: A Comparative Study of Pattern Theory and C4.5

摘要

The Machine Learning field has identified several different inductive biasclasses with Occam's Razor being held as an accepted paradigm. C4.5, an extension of ID3, is one of the leaders in this class of learning systems with which other systems measure their ability. A completely different approach, yet still a method in the class of Occam biased learning mechanisms, is Pattern Theory. This approach seeks to recognize patterns in a robust manner using function decomposition. FLASH, the embodiment of Pattern Theory is itself, an inductive learning system. In this study, we hope to show that the Pattern Theoretic approach is not only as good as the classic decision tree methods, but also it exhibits strong promise to be a robust technique to identifying patterns. We will compare C4.5 and Pattern Theory against a special benchmark set of patterns intended to illustrate many types of potential concepts to be learned. The comparisons will be made by constructing learning curves for each system. Machine learning, Patterson theory, Supervised learning, C4.5, Occam-based learning.