MODEL SELECTION IN NEURAL NETWORKS BY USING INFERENCE OF F. INCREMENTAL, PCA AND SIC CRITERION FOR TIME SERIES FORCASTING
| Main Authors: | Suhartono, Suhartono, Subanar, Subanar, Suryo , Guritno |
|---|---|
| Format: | Article PeerReviewed application/pdf |
| Bahasa: | eng |
| Terbitan: |
, 2006
|
| Subjects: | |
| Online Access: |
https://repository.ugm.ac.id/32902/1/1.pdf https://repository.ugm.ac.id/32902/ |
Daftar Isi:
- Abstract. The aim of this paper is to discuss and propose a procedure for model selection in neural network for time series forecasting. We focus on the model selection strategies based on statistical concept, particularly on the inference of R2 incremental, Principal Component Analysis (PCA) of the residual model and SIC criterion. In this paper, we employ this new procedure in two main approaches for model selection in neural networks, those are bottom-up or forward approach which starts with a large neural networks and top-down or backward approach which begins with an empty model. We use simulation as case study. The result show that statistical inference of R2 incremental combined with SIC criterion is an an effective procedure for model selection in neural networks for time series forecasting. Key words: Neural network, model selection, statistical inference, time series forecasting