Swarm Based Ensembles for Time Series Residual Forecasting

The forecasting of residual series has become a promising approach to improve the accuracy of forecasting systems. One important aspect of residual forecasting is to employ a proper model selection method, in order to reduce model uncertainty and improve the accuracy. In this work, an ARIMA model is...

Full description

Saved in:

Bibliographic Details
Published in:	2020 IEEE 32nd International Conference on Tools with Artificial Intelligence (ICTAI) pp. 595 - 602
Main Authors:	de Holanda, Rafael Bergamo B., de Oliveira, Joao F. L.
Format:	Conference Proceeding
Language:	English
Published:	IEEE 01-11-2020
Subjects:	Forecasting Genetic algorithms Optimization Particle Swarm Optimization Predictive models Sociology Support Vector Regression Time series analysis Time series Forecasting Uncertainty
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	The forecasting of residual series has become a promising approach to improve the accuracy of forecasting systems. One important aspect of residual forecasting is to employ a proper model selection method, in order to reduce model uncertainty and improve the accuracy. In this work, an ARIMA model is used to perform linear forecasts and the residuals are modelled by a support vector regression optimized by a particle swarm optimization algorithm in order to reduce the errors from model misspecification, and the combination of the population is also used to reduce the errors from model variance through average and median operators. Furthermore, in order to avoid problems from the optimization, a subset of the population is selected and combined to perform forecasts. The experiments were performed on six real world time series, comparing with hybrid systems presented in the literature. The results show that the proposed method with the selection of the subset of the population achieved the best overall results.
ISSN:	2375-0197
DOI:	10.1109/ICTAI50040.2020.00097