Comparison of different algorithms and processing techniques for time series classification of academic emotions using EEG data

Use of brainwave signals recorded via Electroencephalogram (EEG) to determine the emotions of a subject is currently established in existing research. Identification of the emotional or affective state of students specifically, is also done to better engage them into learning. As data, EEG signals are time-bound since the frequency and intensity of the signals tend to change over time. The affective state of students too can be considered time-bound since they are expected to vary over time as they engage in learning. This research thus used time series analysis techniques on the EEG readings of academically achieving high school students while they engage in learning through computer based learning systems to determine their affective state with the aim of identifying patterns in the data over time. The use of Discrete Wavelet Transform to include time bound information to the analysis of the EEG data was able to train classifiers with decent results which marginally improved classification performance compared to the precursor study of this paper. Symbolic strings were transformed from the same data used for classification using SAX and a modified Shapelet Transform. These strings were shortened to ease analysis and then patterns within the strings transformed from the same feature and emotion were observed which differ significantly to strings transformed from different features and emotions. These results show that there is promise in considering the temporal aspect of emotion recognition