Locality preserving projections in affective electroencephalogram analysis

Abstract

This work investigates the feasibility of Locality Preserving Projections (LPP) for emotional classification to explore whether any advantages may be obtained from the characteristics of the LPP. Electroencephalogram (EEG) signals obtained from the DEAP: A Database for Emotion Analysis using Physiological Signals were used. The frequency power of trials using Welch’s method were computed and averaged over the theta, alpha, beta and gamma bands. Correlations between the EEG power of frequency over the 4 bands and the participants’ ratings was investigated. Data is of high dimensionality and thus a dimension reduction algorithm was proposed. The LPP was proposed and used in this project. Methods and results are presented for single trial classification and a classifier is built for each person. The results of dimensionality reduction from the LPP were compared to the results obtained from feature selection. Feature selection methods similar to those found in the literature were computed. Only a small improvement in performance was noted with the LPP over feature selection. Finally, it was also noted that the EEG data in 2 dimensions is not isomorphic to the location of the participants’ ratings.