Range Automata for Alphabetic Time Series Dimensionality Reduction

This paper presents a symbolic time series representation technique using range automata model with dimensionality reduction in large datasets. The proposed technique uses automata model to convert the numeric time series into sequence of symbols from which useful information can be extracted. The important task is to define the number of symbols required for time seriesrepresentation; to identify the cutoff value for each symbol and construct range automata for converting the numeric time series into symbolic one dynamically. The present work addresses the above issues by considering the amplitude range of the time series. The proposed method is validated by applyingit on the time series data of ECG, which proves it to be an effective method for time series representation. The performance of RATSR is compared with SAX for elapsed time for conversion and number of samples in reduced dimension.