Emotion Detection by Electroencephalogram

Abstract

Emotion Recognition is probably one of the biggest challenges engineers are facing now. With ongoing development and demand of man - machine interface it has been very necessary to implement emotion recognition. But emotion not being standardized or quantifiable, it is very difficult to classify. Again the lack of proper benchmark for differentiating between various emotions makes it a more difficult challenge. Again the outcome of emotions can be noticed in EEG signals - which again are very hard to classify. According to biomedical science, the EEG data can be classified in not amplitude domain, but in frequency domain - thus making the work more challenging. This projects uses machine learning with other statistical variations like PCA to find the benchmark for emotions of persons having similar behavioural characteristics and classify the emotions. The first stage of the work was detection of Epilepsy and classify the signals into epileptic and non-epileptic.This task hes been achieved by using conventional statistical moments like mean and variance. On a later stage machine learning was applied to classify. The later part of work was to use nonlinear SVM and get emotion data for training and classification. A number of emotions like valence, arousal, dominance etc. Further the non-linear classification algorithm was tested on several dataset. The final stage of work includes HDL coding for implemetation of non-linear SVM.