A Time Series Data Classification Method Based on Basis Function Expansion and Bagging Framework

For the time series data classification problem, we propose a method combining basis function expansion and Bagging framework. The proposed method can capture the functional features of time series by using basis function expansion, and deal with the problems of high dimension and large noise of time series. Furthermore, since the importance of each coefficient of basis function expansion for the response variable is unknown, we can model it using a bagging framework. Since the base classifier is free, we can use different classification models to capture the linear and nonlinear information between the coefficients of basis function expansion and the response variable. In the actual data analysis, we consider three expansion methods such as functional principal component basis, Fourier basis, and B-spline basis, and use GBDT as the base classifier under the Bagging framework. The experimental results show that the proposed method has better classification performance than the traditional classification method and shows a certain universality.