Historical time series prediction framework based on recurrent neural network using multivariate time series

Abstract

In this research, I propose a method for predicting future time series using multivariate historical time series. Historical time series refers to information that varies with time, such as stock prices and economic indicators, and is distinguished from physical time series like voice. Historical time series differs from physical time series of origin in that multiple factors are structured as a spider thread-like interactions. It is not possible to link the causal relationship between these factors, and this is a major aspect that makes historical time series prediction difficult. In this research, a framework for statistically solving historical time series prediction was devised using a deep learning method and its usefulness was confirmed experimentally.