City Research Online

The FL-SMIA Network: A Novel Architecture for Time Series Prediction

Mahdi, A., Weyde, T. ORCID: 0000-0001-8028-9905 and Al-Jumeily, D. (2018). The FL-SMIA Network: A Novel Architecture for Time Series Prediction. 2017 10TH International Conference on Developments in eSystems Engingeering (DeSE), doi: 10.1109/DeSE.2017.42

Abstract

In this paper we propose the FL-SMIA model, a novel neural network model that combines the principles of the Functional Link Neural Network (FLNN) with the Self-organizing Multilayer Neural Network using the Immune Algorithm (SMIA). We describe the FL-SMIA architecture and operation and evaluate its predictive performance on different financial time series in comparison to other neural network models. The FL-SMIA model combines the higher-order inputs of the tensor-product FLNN, i.e. the products of raw input features, with the self-organizing hidden layer of SMIA that dynamically grows and adapts to the input vectors. The FL-SMIA has two advantages over other models. First, it can dynamically adapt to growing amounts of data with a model that grows increasingly complex. Second, it keeps an explicit representation of the patterns it recognises in the data. Experimental results show that the FL-SMIA improves performance, as measured by annualised return in five-days-ahead and one-day-ahead prediction tasks for share prices and exchange rates, over the SMIA networks alone and over standard multilayer perceptrons. It performs on the same level as the FLNN, sometimes better but not significantly so. The result that FLNN and FL-SMIA outperform other multilayer models indicates that particularly the higher-order features contribute to the improved performance and motivate further research into mixed neural network architectures for financial time series prediction.

Publication Type: Article
Additional Information: © 2018 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.
Departments: School of Mathematics, Computer Science & Engineering > Computer Science
URI: http://openaccess.city.ac.uk/id/eprint/19769
[img]
Preview
Text - Accepted Version
Download (325kB) | Preview

Export

Downloads

Downloads per month over past year

View more statistics

Actions (login required)

Admin Login Admin Login