Fitting Optimal Classification Trees Using an Elitist Genetic Algorithm with Bernoulli Crossover
Loading...
Date
2020-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Optimal Classification Trees (OCTs) and Optimal Regression Trees (ORTs) promise to provide empirical modeling methods with unparalleled combinations of accuracy and interpretability. Yet, the computation times of the fitting procedures, which are currently based on integer programming solvers and branch and cut methods, tend to grow exponentially with the problem size. Therefore, problems involving over 3,000 datapoints may be practically out of reach. The purpose of this thesis is to explore the feasibility of Genetic Algorithms (GAs) to efficiently solve classification problems using variants of the well-known Iris Data classification problem. By studying this problem, it is verified that GAs can solve the problem optimally within a few seconds. Also, the GA time grows linearly with respect to the number of generations, the number in the population, and the number of runs. Therefore, exploring GAs to solve OCTs and ORTs is a promising topic for future research.
Description
Keywords
Genetic Algorithms, Classification Trees, Computer Experiments