Multi-view kernel PCA for time series forecasting

In this paper, we propose a kernel principal component analysis model for multi-variate time series forecasting, where the training and prediction schemes are derived from multi-view formulation of Restricted Kernel Machines. The problem is simply an eigenvalue decomposition summation two matrices corresponding to views input output data. When linear used view, it shown that forecasting equation takes form ridge regression. non-linear, pre-image has be solved forecast point in space. We evaluate on several standard datasets, perform ablation studies, benchmark with closely related models discuss its results.