DATA MINING OF MULTIPLE NONSTATIONARY TIME SERIES
Richard J. Povinelli, Xin Feng · 1999
A data mining method for synthesizing multiple time series is presented. Based on a single time series algorithm, the method embeds multiple time series into a phase space. The reconstructed state space allows temporal pattern extraction and local model development. Using an a priori data mining objective, an optimal local model is chosen for short-term forecasting. For the same sampling period, multiple time series embedding produces better temporal patterns than single time series embedding. The method is applied to a financial time series.