Abstract
The using of artificial neural network (ANN) in past decade has led many researchers adapt ANN to do forecasting routines. Back-propagation network (BPN) is the most popular network algorithm because it is easier to understand and its convergence is better. However, BPN still has many weaknesses in its long learning time and local minima problem. The objective of this project is to research on the effectivenes of back-propagation algorithm in time series data forecasting. To evaluate the performance of standard back-propagation algorithm, the forecast result from each Krzyzak algorithm and improve back-propagation algorithm is used for comparison. The algorithms are compared under two aspects which are convergence and learning time. C programming language is used for this research. In the end of the research. Improve back-propagation algorithm is found to be better than standard back-propagation algorithm and Krzyzak algorithm in terms of convergence and learning time. Improve back-propagation algorithm requires simple structure for learning, but showed negative result when the maximum error is low. Krzyzak algorithm's convergence is satisfying and requires shorter time for learning when compared to standard back-propagation algorithm. Besides that, Krzyzak algorithm's convergence improves when the maximum error is low.
Website : http://ir.fsksm.utm.my/1520/
No comments:
Post a Comment