An improved Pi-Sigma neural network using error feedback for time series prediction

Akram, Urooj (2018) An improved Pi-Sigma neural network using error feedback for time series prediction. Masters thesis, Universiti Tun Hussein Onn Malaysia.


Download (520kB) | Preview
[img] Text (Copyright Declaration)
Restricted to Repository staff only

Download (492kB) | Request a copy
[img] Text (Full Text)
Restricted to Registered users only

Download (1MB) | Request a copy


Time series prediction grabs much attention because of its effect on the vast range of real-life applications. Traditional time series forecasting tools have some limitations like slow training process, less efficient training methods that decrease the performance of the model. Higher Order Neural Network (HONN) using recurrent feedback appeared as a powerful technique in the domain of time series prediction and it has the ability to expand the input space, making them more efficient for solving complex problems and perform high learning abilities in time series prediction. This study proposed a model called improved Pi-Sigma Neural Network using Error Feedback (PSNN-EF) which combines the properties of Pi-Sigma Neural Network (PSNN), recurrence and error feedback. PSNN-EF uses backpropagation gradient descent algorithm for training purpose and is tested with physical time series signals of humidity, evaporation and wind direction datasets that are collected from Malaysian Meteorological Department (MMD). The prediction result is compared with Jordan Pi-Sigma Neural Network (JPSN) and the ordinary PSNN. The results clearly showed that the PSNN-EF significantly improved the computational efficiency of the training process and has been developed to produce more realistic and acceptable results. The average improvement of the proposed model on evaporation dataset is 2.06%, humidity is 7.45% and wind is 3.51% as compared to other models. The benefit of using error feedback is that it generates more accurate and promising results of prediction. Therefore, from the performance of the proposed method, it is noticed that PSNN-EF can provide better solution to JPSN for one-step-ahead prediction of those three datasets.

Item Type: Thesis (Masters)
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75-76.95 Calculating machines
Divisions: Faculty of Computer Science and Information Technology > Department of Software Engineering
Depositing User: Miss Afiqah Faiqah Mohd Hafiz
Date Deposited: 25 Jul 2021 08:41
Last Modified: 25 Jul 2021 08:41

Actions (login required)

View Item View Item