Trends by adaptive fourier decomposition and application in prediction

Abstract

In this paper, we introduce the concept of trend function made from adaptive Fourier decomposition. Through applying a data preprocessing method, we achieve trend functions of the rational function type. The trend functions are divided into different levels to describe the tendency of the signal or time series in question. They are, in particular, smooth, and are free from the Gibbs phenomenon as usually introduced by the Fourier type methods. To justify the role of the trend function concept, an interpolation result is proved. An effective application of trend function in predicting stochastic time series is presented based on real COVID-19 outbreak data of China. The algorithm aspect is fully discussed.