Time Series Forecasting using Wavelets with Predictor-Corrector Boundary Treatment (2001)
| Venue: | in 7th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining |
| Citations: | 6 - 1 self |
BibTeX
@INPROCEEDINGS{Yu01timeseries,
author = {Peng Yu and Anna Goldenberg and Zhiqiang Bi},
title = {Time Series Forecasting using Wavelets with Predictor-Corrector Boundary Treatment},
booktitle = {in 7th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
year = {2001}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Abstract Time series analysis has attracted a lot of attention due to its applicability in a wide range of fields. Various methods, such as linear and nonlinear models, and wavelets have been used for time-series analysis and predictions. We propose a new methodology that can be integrated with existing techniques to iteratively achieve better accuracy. We decompose the original signal into several components using stationary wavelet transform and forecast each component with a linear model. Predictions from all components are summed up to give the overall prediction. Our method, the predictor-corrector boundary treatment, is a new heuristic approach to solving the boundary problem. Rather than using an adhoc condition, we use the predicted values from the previous step for the boundary of the wavelet transform and then iterate. Preliminary experiments have shown that the algorithm improves its predictions with every iteration and is a promising methodology for time series analysis.
Keyphrases
predictor-corrector boundary treatment time series forecasting adhoc condition original signal various method wide range time-series analysis overall prediction time series analysis new heuristic approach abstract time series analysis predicted value nonlinear model linear model stationary wavelet transform new methodology several component boundary problem previous step promising methodology preliminary experiment wavelet transform
