A modified weight optimisation for higher-order neural network in time series prediction

Husaini, Noor Aida (2020) A modified weight optimisation for higher-order neural network in time series prediction. Doctoral thesis, Universiti Tun Hussein Onn Malaysia.

[img]
Preview
Text
24p NOOR AIDA HUSAINI.pdf

Download (13MB) | Preview
[img] Text (Copyright Declaration)
NOOR AIDA HUSAINI COPYRIGHT DECLARATION.pdf
Restricted to Repository staff only

Download (13MB) | Request a copy
[img] Text (Full Text)
NOOR AIDA HUSAINI WATERMARK.pdf
Restricted to Registered users only

Download (14MB) | Request a copy

Abstract

Most of time series signals are difficult to predict as consist of non-linear, high complexity (noise) and chaotic processes. The challenges in time series prediction are to provide a technique to better understand a dataset. In line with this, the Cuckoo Search (CS) learning algorithm, a kind of metaheuristics techniques employs high-level techniques for exploration and exploitation of the search space in which its step length is much longer in the long run. Thus, can explicitly being used to address the possibilities of stochastic trends in time series signals. Since its discovery, the CS has been used extensively. However, these methods fixed the parameter values which essential for adjusting the weights. Therefore, a modification was made by the additional step of information exchange between the top eggs, which significantly improve the convergence rate. Hence, motivated by the advantages of those Modified Cuckoo Search (MCS), the improvement of the MCS called Modified Cuckoo Search-Markov chain Monté Carlo (MCS-MCMC) learning algorithm is proposed for weight optimisation. As the Markov chain Monté Carlo can replace the cumbersome in generating the objective functions, it is used to substitute the Lévy flight found in the MCS’s structure to prove that MCS-MCMC is suitable for predictive tasks. The performance of MCS-MCMC learning algorithm was validated with several test functions and compared with those of MCS learning algorithm. The MCS-MCMC results is further benchmarked with the standard Multilayer Perceptron, standard Pi-Sigma Neural Network (PSNN), Pi-Sigma Neural Network-Modified Cuckoo Search, Pi-Sigma Neural Network-Markov chain Monté Carlo, standard Functional Link Neural Network (FLNN), Functional Link Neural Network-Modified Cuckoo Search and Functional Link Neural Network-Markov chain Monté Carlo which emphasis in optimising the accuracy rate. The simulation results proved that MCS-MCMC outperformed in the form of Accuracy with the range of 0.003% to 4.421% when incorporated with standard PSNN and FLNN for three (3) data partitions covering 10 benchmarked time series datasets.

Item Type: Thesis (Doctoral)
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines
Divisions: Faculty of Computer Science and Information Technology > Department of Software Engineering
Depositing User: Mrs. Sabarina Che Mat
Date Deposited: 09 Sep 2021 04:02
Last Modified: 09 Sep 2021 04:02
URI: http://eprints.uthm.edu.my/id/eprint/921

Actions (login required)

View Item View Item