Prediction of Chaotic Time Series

This research deals with a problem of chaotic time series prediction. Using the concepts of non-predictable points, predictive clustering and motifs for time series prediction has shown impressive results in the prior research on this topic. This research introduces a new concept: base and auxiliary time series. Base time series is the one that is desired to predict, auxiliary time series is used for training the algorithm and is not being predicted. Also an auxiliary set has been introduced - a set of auxiliary series. As a result of the research, the optimal size of the auxiliary set has been estimated. Moreover, an introduction of the auxiliary set has shown that the percent of non-predictable points grows more slowly compared to an original algorithm and the error rate remains acceptable. Furthermore, the length of used time series was smaller and the results of predictions did not suffer.