An intelligent approach to time series identification by a neural network-driven decision tree classifier

The objective of this paper is to suggest a new intelligent approach to classifying a time series into one of autoregressive moving-average (ARMA) models, which is named time series identification (TSI), by using a neural network-driven decision tree classifier, The main recipe of our approach is to apply two pattern recognition concepts for solving the TSI problem. The first pattern recognition concept is an extended sample autocorrelation function which is derived from a given times series data and is used as an important feature for solving the TSI problem. The second pattern recognition concept is a neural network-driven decision tree classifier which is a main vehicle for reducing the complexities involved in TSI problems and, finally, providing the most promising ARMA model for a given time series. The neural network-driven decision tree classifier consists of a set of nodes at which neural network-driven decision making is made whether the connecting subtrees should be pruned or not. To enhance the performance of our proposed classifier, we suggest a neural pruning search algorithm which is used to find the promising paths. The proposed search algorithm essentially results in a neural network-driven search through the space of possible terminal nodes of the classifier. Experimental results with a set of real time series data show that the proposed approach can efficiently identify the time series patterns with high precision compared to other approaches.