Uso de meta-aprendizado para avaliar a tunagem para o algoritmo Máquina de vetores de suporte

Abstract

Machine learning is one of the fields of study in the area of Artificial Intelligence, which uses algorithms that perform classification and clustering of data to extract knowledge to solve a given problem. Machine Learning algorithms use hyper-parameters to improve their performance, and these hyper-parameters are adjustable to adapt to the problem. The search for these hyper-parameters can be quite complex, especially when applied to complex databases. This work uses the meta-learning technique to predict if there is a real advantage of improving the hyper-parameters for the Support Vector Machine algorithm using the Random Search, Grid Search, Genetic Algorithm, and Particle Swarm Optimization. The results indicate that Random Search algorithm has satisfactory results on non-complex bases, and on complex bases, the default hyper-parameters have better efficiency., and for the results of using meta-learning, the results show that they had the best relationship with the baseline and improve the results when the meta-learning was fine-tuned.