Study of the applications of symbolic regression in real world data sets

• Grant number: 17/17815-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: November 01, 2017
• End date: August 11, 2019
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Fabricio Olivetti de França
• Grantee: Daniel de Pina Nascimento
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil

Abstract

Many decision process are modeled as a regression problem. Given a data set, it is expected to obtain a numeric response. Basically, we use Linear Regression when we wish to obtain a simple model easy to interpret, on the other hand, many data sets express a nonlinear relationship with the target variable, and so nonlinear regression models are used. Many known nonlinear regression models are defined as a composition of a chosen nonlinear function in order to approximate the target function. But, such approach leads to models difficult to interpret known as black box models. Symbolic Regression on the other hand, search for nonlinear relationship with the objective of minimizing the approximation error while maximizing the simplicity of the generated expression The most well known algorithm used to optimize such model is Genetic Programming that, despite its success, it still considered to be distant from the desirable results. Many other techniques were proposed to try to balance those two objectives. This project has the goal to compare state-of-the-art Symbolic Regression techniques in real world data sets with the objective of comparing their performance. (AU)