Um estudo comparativo entre algoritmos para otimização de hiperparâmetros

Abstract

Artificial intelligence and machine learning models have become increasingly relevant over the years, because of the large amount of applications that they have in most areas of society. Therefore, the study of ways to improve the quality of these systems becomes very relevant, and one of the ways to improve their performance is through the hyperparameters optimization. Thus, this work aims to analyze comparatively the application of different algorithms in the hyperparameter optimization of machine learning models. Five optimization techniques are discussed, these being, grid search, random search, bayesian search, genetic algorithm and particle swarm optimization, where a research is accomplished about solutions for integration of these techniques aiming to facilitate their application for the optimization of either the addressed models in this work or other learning models. The comparative analysis is done by exploring supervised learning tasks, using for classification databases and one regression database, with the objective of comparing the efficiency of each optimization technique in different problems and also comparing the influences of specific features of each model, such as the number of hyperpameters on each one.