Random Fourier feature kernel recursive maximum mixture correntropy algorithm for online time series prediction.

Area of Science:

• Machine Learning
• Signal Processing
• Time Series Analysis

Background:

• Kernel Recursive Least-Squares (KRLS) algorithms are widely used for prediction tasks.
• Existing KRLS methods can suffer from high computational complexity and limited robustness.
• There is a need for improved algorithms that balance efficiency and accuracy in time series prediction.

Purpose of the Study:

• To propose a novel algorithm, the Random Fourier Feature Kernel Recursive Maximum Mixture Correntropy (RFF-RMMC), to enhance KRLS performance.
• To improve prediction efficiency and robustness in online time series prediction.
• To address the trade-off between computational cost and prediction accuracy in kernel-based methods.

Main Methods:

• Approximation of kernel functions using Random Fourier Features (RFF) to reduce computational complexity.
• Integration of the Maximum Mixture Correntropy Criterion (MMCC) to enhance robustness and accuracy.
• Application of the combined RFF-MMCC approach within the KRLS framework for online learning.

Main Results:

• The RFF-RMMC algorithm significantly reduces computational complexity compared to traditional KRLS.
• The MMCC criterion enhances the accuracy of similarity measurements, compensating for potential accuracy loss from RFF.
• Simulation results on three datasets demonstrate the improved prediction efficiency and robustness of the proposed RFF-RMMC algorithm.

Conclusions:

• The RFF-RMMC algorithm offers a superior alternative for online time series prediction by effectively combining RFF and MMCC.
• The proposed method achieves a favorable balance between computational efficiency and prediction accuracy.
• This work contributes a robust and efficient kernel-based prediction algorithm for various applications.