Selección de entradas y salidas en DEA mediante algoritmos genéticos multiobjetivo

Abstract

ABSTRACT : This study presents a method for feature selection in datasets that suffer from high dimensionality, aimed at evaluating relative efficiency using the nonparametric technique known as Data Envelopment Analysis (DEA). This application is implemented using genetic algorithms for multi-objective optimization, applying a basic binary structure to select the variables from the dataset that will enter the DEA evaluation. The selection process is carried out through crossover and mutation of an initially randomly selected population from the solution space. Four different well-known multiobjective genetic algorithms are used and compared using data sets from the DEA literature.