Brain-Computer Interface (BCI) based emotion recognition is mostly motivated by the supervised technique and may get biased by the emotion category representation. An alternative approach to emotion recognition could be employing an unsupervised technique. We compared a supervised and an unsupervised technique for two-class emotion recognition using time-frequency-based EEG features. Results obtained from the unsupervised technique reveal incongruency between the actual physiological response and the established ground truth (emotion category), whereas an average accuracy of 62.76% is observed for different emotion categories while employing the supervised technique.