The aim of this study is to select the best features of EEG signal, by in-vestigating the AdaBoost feature importance score measure as a means to find a ranking of important features which can improve the classifier performance for recognizing the imagined speech of 8 Indonesian words, i.e., makan (eat), minum (drink), lapar (hun-gry), haus (thirsty), senang (happy), sedih (sad), sakit (sick) and toilet (toilet). The EEG signal was recorded from 11 healthy students, 7 men and 4 women, using Emotiv epoch and Emotiv Pro. Feature importance score was applied to AdaBoost model.
Our research showed that the top ten features based on feature importance score ranking of AdaBoost model were T7 GAMMA, T7 THETA, P7 HIGH BETA, P8 GAMMA, P8 HIGH BETA, F3 GAMMA, F3 HIGH BETA, T7 HIGH BETA, P7 GAMMA and FC5 THETA, with the resulting accuracy 75%, precision 80% and sensitivity or recall 84%.