The NN-Stacking: Feature weighted linear stacking through neural networks

Author(s)

Coscrato, Victor ^{[1, 2]} ; de Almeida Inacio, Marco Henrique ^{[1, 2, 3]} ; Izbicki, Rafael ^[1]

Affiliation

^[1] Univ Fed Sao Carlos, Rodovia Washington Luis, Km 235, Sao Carlos, SP - Brazil

^[2] Univ Sao Paulo, ICMC, Av Trabalhador Sao Carlene, 400 Ctr, Sao Carlos, SP - Brazil

^[3] BME, TMIT, Muegyet Rkp 3, H-1111 Budapest - Hungary

• Document type: Journal article
• Source: Neurocomputing; v. 399, p. 141-152, JUL 25 2020.
• Web of Science Citations: 0

Abstract

Stacking methods improve the prediction performance of regression models. A simple way to stack base regressions estimators is by combining them linearly, as done by Breiman {[}1]. Even though this approach is useful from an interpretative perspective, it often does not lead to high predictive power. We propose the NN-Stacking method (NNS), which generalizes Breiman's method by allowing the linear parameters to vary with input features. This improvement enables NNS to take advantage of the fact that distinct base models often perform better at different regions of the feature space. Our method uses neural networks to estimate the stacking coefficients. We show that while our approach keeps the interpretative features of Breiman's method at a local level, it leads to better predictive power, especially in datasets with large sample sizes. (C) 2020 Elsevier B.V. All rights reserved. (AU)

• FAPESP's process: 19/11321-9 - Neural networks in statistical inference problems
• Grantee: Rafael Izbicki
• Support Opportunities: Regular Research Grants
• FAPESP's process: 17/03363-8 - Interpretability and efficiency in hypothesis tests
• Grantee: Rafael Izbicki
• Support Opportunities: Regular Research Grants