Unified Feature Selection and Hyperparameter Bayesian Optimization for Machine Learning based Regression

摘要

In this paper we propose a method that is based on Bayesian optimization that performs feature selection and hyperparameter optimization for training regression models. Without loss of generality, we consider here as regressor a Multilayer Perceptron neural network (MLP) and as feature selection criterion the distance correlation measure. The performances of the proposed method are analyzed when fitting data generated by a custom function with known dependence of variables and some real data with unknown dependence.