Effect of dimension reduction on predictability of multivariate chaotic time series

다국어 초록 (Multilingual Abstract)

Dimension reduction is an important component of a machine learning area. It transforms input spaces into the reduced spaces with smaller dimensionality. Goal of this paper is to analysis the effect of using various dimension reduction techniques for predicting multivariate chaotic time series. Input space of multivariate chaotic time series which is reconstructed state space usually brings more information of an original strange attractor than one of univariate chaotic time series. When the multivariate chaotic time series are used, however, it exhibits relatively high dimension on time delay coordinates vector which induces curse of dimensionality, statistical dependency and redundancy among features of input spaces which disturb the ability of machine learning techniques. To solve this problem, we apply dimension reduction techniques. After that, least squares support vector regression (LSSVR) of machine learning techniques is used to predict future value of chaotic time series. Our experiment consists of delayed Lorenz series.

목차 (Table of Contents)

• Abstract
• 1. Introduction
• 2. Chaotic time series
• 3. Dimension reduction
• 4. LSSVR (Least squares support vector regression)
• 5. Experiment
• 6. Conclusion
• 7. Acknowledgements
• 8. Reference