Ferramenta computacional para apoio à análise de regressão de dados

Abstract

Regression analysis has application in several areas, and the models obtained can be used afterwards in optimization processes. Choosing the best model for a given databases, however, is still a time-consuming task, which is frequently done in a heuristic way. In this work we develop a computational tool to support the choice of models and the estimation of parameters, based on the application of a Genetic Algorithm to the three groups of models: Least Mean Squares ones, Multi Layer Perceptrons and k-Nearest Neighbors ones. The tool is tested in different databases, one of them originating from the area of structure-based rational drug design, where future applications are foreseen.