Analysing Time Series Data

Abstract

This paper investigates the use of auditory perceptualisation for analysing the statistical properties of time series data. We introduce the problem domain and provide basic background on higher order statistics like skewness and kurtosis. The chosen approach was direct audification because of the inherent time line and the high number of data points usually available for time series. We present the tools we developed to investigate this problem domain and elaborate on a listening test we conducted to find perceptual dimensions that would correlate with the statistical properties. The results indicate that there is evidence that kurtosis correlates with roughness or sharpness and that participants were able to distinguish signals with increasing difference of the kurtosis. For the setting in the experiment the just noticeable difference was found to be 5. The collected data did not show any similar evidence for skewness and it remains unclear whether this is perceivable in direct audification at all.