Forecasting Model over Random Interval Data Stream Based on Kalman Filter
Yu Yonghong, Bai Wen-yang · 2009
It is very important in a lot of applications to forecast future trend of data streams. Recent works on prediction analysis over data streams mainly supposed that data are complete and data occur at equal time interval. Adopting state transition of time series and Kalman filter, a predictive model for forecasting the trend of data stream with missing values and data occurring in random time interval is proposed in the paper. The proposed model adopts Kalman gain matrix to compute automatically the maximum likelihood estimation of data stream to obtain optimal estimates in linear, no deviation, and minimum mean square error way. Experiment shows that the proposed model has higher performance and provides better trend prediction of data stream in bounded memory and limited run time, and it can predict future trend of data streams online.