A New Approach Of Constructing Decision Tree Based On Shortest Path Methods

Decision tree is a commonly used machine learning classification method. When generating a decision tree, it is always in the direction of reducing information entropy. Hence, the construction of decision tree has a lot in common with path planning. In recent years, many scholars have proposed methods using path planning solutions to solve data classification problems. The most notable is the Ant-Miner algorithm proposed by Parepinelli, Lopes, and Freitas in 2002. These algorithms have seen good performance in classification problems, however, when taking calculation steps into account, they appear to be sort of complicated. To make the path-planning-based algorithm more concise and comprehensible, we propose a new algorithm, abstracting the dataset to a directed graph, using common path planning algorithm to find an optimum solution, and finally forming a decision tree.