Decision support in time series modeling by pattern recognition

This research is aimed at presenting a new, pattern recognition-based DSS scheme for the time series model identification. The scheme is based on two principles: pattern matching and inductive learning. Pattern matching is used to classify a pattern of the time series into one of the autoregressive moving-average models. The pattern is obtained from the extended sample autocorrelations of the time series. Inductive learning is used to enhance the capability of recognizing input patterns, and linear discriminants are used to discriminate one pattern from the others. To implement the idea, a decision support system named DSSTSM was designed and a prototype was developed on the microcomputer. Experimental results show that the combination of the pattern recognition principles with a DSS can yield a promising solution to the time series modeling.