A Time Series Regression Model Based on Distance Covariance Weighted Bagging Framework and Functional Principal Component Analysis

For the regression prediction problem of time series data, we propose a method based on functional principal component analysis and distance covariance weighted Bagging framework. The proposed method overcomes the problems of high dimension and noise of time series data, and preserves the function characteristics of time series. Since the underlying model structure between the principal component basis coefficient variable and the response variable is unknown, we use the Bagging framework, which is with the model-free basis model, to capture the potential model, considering that the importance of each basis model is unknown, we propose a method based on distance covariance weighting to correct the Bagging framework to improve the prediction accuracy.