Time series prediction with missing values: a bidirectional fully connected neural network approach

Time series prediction with missing values is a universal problem in various fields, while existing models cannot satisfy the requirements of engineering systems. To solve this challenge, an effective model is proposed: Bi-FCNN, which combines the bidirectional fully connected neural networks (FCNN) and the back propagation (BP) algorithm with linear propagation and nonlinear mapping capabilities, and performs well in temporal information with missing values forecasting tasks. Analysis of comparative experiment indicates that Bi-FCNN could be competent in missing data environment for predicting deletion that is inescapable due to natural or artificial factors.