Short-Term Multimedia Exposure Estimation from Pupil Dilation: Impact of Normalization

This paper aims to compare different methods of normalizing pupil dilation signals with the purpose of using them for the estimation of short-term multimedia exposure. Broader research, part of which is this paper, aims to classify multimedia exposure based on psychophysiological signals using machine learning. In this paper, the reactance dimension of the multimedia exposure scale is classified using electrodermal activity and pupil dilation signals. The reactance dimension represents negative emotions connected to multimedia exposure. Three different normalization methods are used, one that preserves pupil dilation in units of length, one that preserves pupil dilation in percentage change and one that is designed to enforce the same variance as well as the mean in each subject’s signal. Electrodermal activity is used with pupil dilation, as we require information from both signals to achieve good classification results. We used different machine learning methods and concluded, that the method that enforces the same variance as well as the mean in each subject’s signal proved to be the most successful.