Otimização bayesiana de modelos de ensemble e decision tree learning: Uma contribuição para predição de energia em contextos industriais

Abstract

This article addresses Bayesian optimization for hyperparameterization of ensemble and decision learning models, with consecutive simulations based on initial hyperparameter values, and allowing the determination of a range of hyperparameters to be inserted and evaluated by the Optuna Bayesian optimization framework. The proposed method is used to predict energy consumption every 15 minutes in a metallurgical industry. The Decision Tree Regressor, Random Forest Regressor and Cubist Regressor models were used, whose performance was previously evaluated. The results obtained surpassed the performances recorded for the same state-of-the-art models in the proposed context.