Multi-step Prediction for Time Series with Factor Mining and Neural Network

Abstract

Multi-step prediction for time series is a challenging research area with broad applications which can provide important information for relevant decision-makers. Many works extended different architecture of artificial neural networks to perform time series prediction, but they mostly only consider the time series itself, does not weigh the impact of time series of relevant factors. In this paper, a new method of time series prediction based on factor mining is proposed. By analyzing target time series, the means of discovering factors influencing time series and pinned down the most relevant factors was proposed. In the end, a method to do multi-step prediction with artificial neural networks, MTPF is proposed to conduct the time series prediction, create time series model and forecast time series. The proposed method is applied for a shipping price index time series prediction. Results show that this method can improve accuracy of prediction when compared with traditional methods.