EEG-Based Emotion Classification Using Wavelet Decomposition and K-Nearest Neighbor

Research on the correlation of EEG signals to emotions based on high/low arousal and valence, has been done before. Some research using the Eigen-Emotion Pattern Kernel method and the Support Vector Machine. The others using the Higuchi Fractal Dimension (FD) Spectrum, the Multifractal Detrended Fluctuation Analysis (MDFA) and the Hidden Markov Model (HMM), but the accuracy is not too good. This research uses Wavelet Decomposition and k-Nearest Neighbor to improve accuracy. The results show that the optimum k values of the k-Nearest Neighbor parameters for this research are 21. Valence's classification accuracy results using Wavelet and k-NN, compared with previous research has the same relative accuracy, ie 57.5%. While the result of arousal classification accuracy using wavelet and k-NN is 64.0%, better than previous research.