Enhanced symbolic aggregate approximation method for financial time series data representation

Data representation is one of the most important tasks in time series data pre-processing. Time series data representation is required to make the data more suitable for data mining specifically for prediction. Time series data is characterized by its numerical and continuous values. One of the data representation methods for time series is the Symbolic Aggregate Approximation (SAX) which uses mean values as the basis of representation of the data. However. representing the time series financial data with the mean value often causes the loss of patterns that can describes important pieces of information. The aim of this study is to propose an enhancement of SAX representation purposely for the financial time series data. The Enhanced SAX (EN-SAX) adds two new values to the original mean value for each segment in SAX. These values enable better representation for each segment in a lower dimension and keep some of the important patterns that are meaningful in financial time series data. The experimental results show that the EN-SAX representation manages to give lower error rates compared to SAX and improves the prediction accuracy.