Removal of artifacts is an important step in any research in /application of electroencephalogram (EEG). The artifacts may contain eye-blinking, muscle noise, heart signal, line noise, and environmental effect. Such noises often make the raw EEG signals not very useful for extraction/identification of physiological phenomena from EEG. The independent component analysis (ICA) is a popular technique for artifact removal in brain research and some reports demonstrate that ICA can remove the artifacts with lower (acceptable) loss of information. But, these reports select useful independent components manually, primarily by looking at the scalp-plots. This is of great inconvenience and is a barrier for BCI or real-time applications of EEG. In this paper, we demonstrate that machine learning methods could be quite effective to discriminate useful independent components from artifacts and our findings suggests the possibility of developing a ldquouniversalrdquo machine for artifact removal in EEG.