A new hyperparameter optimization approach for classification models applied to bank account churn data

Abstract

Não inclu resumo

Abstract : In this work we present a new approach for the hyperparameter optimization stage of machine learning classification models applied to banking data. The function, called F3, considers a weighting between the classical optimization function (F1: which considers the maximization of accuracy) and another optimization function that considers the difference between the models’ accuracies (namely F2). Optimization functions were compared for different machine learning classification models, including an ensemble of classifiers, and applied to bank account churn data to highlight the possibilities for companies in the financial sector to prevent losses by predicting and mitigating customer churn in advance. The proposed optimization function presented better results, surpassing the accuracy of the classical approach in the analyzed situations