Ensemble of Deep Belief Network and Bayesian Adaptive Aggregation for Regression

Hassan, Saima; Khanesari, Mojtaba Ahmadieh; Jan, Mohammad Tariq; Khan Mashwani, Wali

Abstract

Ensemble modeling of Neural Networks is a strategy where multiple alternative models (ensemble members) are constructed and then their forecasts are ensembled using various combination approaches. Ensemble of Neural Networks has proved the concept behind this strategy. Deep neural network is a type of neural network that offers potential opportunities to overcome traditional ensemble of neural networks. This paper proposes an ensemble of deep belief networks (DBN). The ensemble members of DBN are constructed with different number of epochs so that the generalization ability can be improved. The outputs of these DBNs are aggregated by a Bayesian model averaging method. The proposed Bayesian adopted ensemble of DBNs is evaluated on two benchmark data sets. Comparison of the proposed model is evaluated with simple averaging and single DBN over a number of forecasting measuring that shows better performance of the proposed model.

• Conference Name: 2019 International Conference on Information Science and Communication Technology (ICISCT)
• Conference Location: Karachi, Pakistan
• Start Date: Mar 9, 2019
• End Date: Mar 10, 2019
• Acceptance Date: Jan 26, 2019
• Online Publication Date: Mar 10, 2019
• Publication Date: 2019-03
• Deposit Date: May 16, 2019
• Publicly Available Date: May 16, 2019
• Pages: 1-6
• Book Title: Proceedings - 2019 International Conference on Information Science and Communication Technology (ICISCT)
• ISBN: 978-1-7281-0448-5
• DOI: https://doi.org/10.1109/cisct.2019.8777443
• Keywords: Ensemble modeling; deep belief network; Bayesian model averaging; forecast combination
• Public URL: https://nottingham-repository.worktribe.com/output/2059083
• Publisher URL: https://ieeexplore.ieee.org/document/8777443
• Related Public URLs: http://uok.edu.pk/icisct/index.html
• Additional Information: © 2019 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.