Evolving decision-tree induction algorithms with a multi-objective hyper-heuristic using the Pareto dominance approach

• Grant number: 15/05218-0
• Support Opportunities: Research Grants - Visiting Researcher Grant - International
• Start date: July 16, 2015
• End date: August 15, 2015
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Márcio Porto Basgalupp
• Grantee: Márcio Porto Basgalupp
• Visiting researcher: Vili Podgorelec
• Visiting researcher institution: University of Maribor (UM), Slovenia
• Host Institution: Instituto de Ciência e Tecnologia (ICT). Universidade Federal de São Paulo (UNIFESP). Campus São José dos Campos. São José dos Campos , SP, Brazil
• Associated research grant: 10/20255-5 - Genetic programming for evolving decision tree induction algorithms, AP.JP

Abstract

The goal of the proposed research collaboration is to design, develop and evaluate an improved decision-tree induction algorithm with a multi-objective hyper-heuristic using the Pareto dominance approach. The previous results have shown the advantages of evolving a decision-tree induction algorithm with a hyper-heuristic regarding the classification performance. However, the classification accuracy and other classification performance metrics are not the sole indicators of the quality of induced decision trees. One very important advantage of decision trees is their transparent representation of knowledge model, which allows a straightforward and simple interpretation that is close to human thinking. It has been shown in many real-world applications, where the validation of the classification results is as important as the classification itself (like medicine, for example), that the simplicity of a decision tree should not be compromised at the expense of accuracy. The both objectives (accuracy and complexity of a decision tree), however, are generally conflicting. To solve this problem, we propose to use a properly designed and implemented multi-objective optimization approach. (AU)