Forays into decision tree learning methods review of methods and tools

Abstract

One of the popular methods for supervised learning is decision trees. It has gained its popularity being a simple to use, easy to understand, and having no need to make any prior assumptions about the data. Decision trees have also achieved veracity of usage as it can be used to construct models for both numerical as well as categorical data. Numerous research studies have been done on the decision trees. This paper aims to study the methods used to construct decision trees, each one having its own significance. This work is focused on surveying these works. It not only covers the popular construction algorithms but also some advanced algorithms and the algorithms present in the commonly used academic research tool. Various induction mechanisms are studied based on the literature from standard publications, well known in the academic and research communities. The methods studied show that the respective algorithms have some pros as well as cons. The algorithms in the tools are tested on a standard dataset to predict the heart disease. The results show that though the same method is used as basis for construction of decision trees in two different tools, the results are quite different. This work, wherein more than 20 algorithms are touched upon and 5 tools are briefed, also helps to understand the overall evolution of the learning strategies for decision tree constructions.