Emotion classification gaining more popular in research world, especially in healthcare. There are many methods can be used to classify emotional state of human and one of them is by using supervised machine learning such as kNN and W-kNN. DEAP's EEG dataset will be used as input signal because of its high dimension. In this research, EEG's non- linear features used to classify the emotional state. We compare recognition rate after variation in feature selection steps to choose which features best uses for this classification. For the emotion classification system we used variation of k parameter in kNN and W-kNN classifier. The results showed that the highest recognition rate was by using Chi-Square selection method with value of 60.15%, but by using those feature selection method did not really give significant difference. Based on that fact, we conclude that DEAP dataset need anoother reliable method to extract its feature and select those feature accurately.