Comparison of Baseline Reduction Methods for Emotion Recognition Based On Electroencephalogram Signals

Emotions play an essential role in human social interactions. Its importance has sparked research on emotion recognition mainly based on electroencephalogram signals. However, differences in individual characteristics significantly affect the electroencephalogram signal pattern and impact the emotion recognition process. Several studies have used the baseline reduction approach with the Difference method to represent the differences in individual characteristics on electroencephalogram signals. On the other hand, the baseline reduction process on signal data, in general, can also use the Relative Difference and Fractional Difference methods. Therefore, the contribution of this research is to compare the performance of the three baseline reduction methods on emotion recognition based on electroencephalogram signals. In this study, feature extraction and representation were also carried out using Differential Entropy and 3D Cube. Furthermore, Convolutional Neural Network and Decision Tree methods are used to classify emotions. The experimental results using the DEAP dataset show that the Relative Difference and Fractional Difference methods are superior in reducing the baseline electroencephalogram signal compared to the Difference method. In addition, the Relative Difference and Fractional Difference methods produce a smoother electroencephalogram signal pattern in the baseline reduction process.