Event-related potentials (ERPs) represent neuronal activity in the brain elicited by external visual/auditory stimulation, and it is widely used in brain-computer interface (BCI) systems. The ERP responses are elicited a few milliseconds after attending to an oddball stimulus; target and non-Target stimulus are repeatedly flashed while the electroencephalography (EEG) is recording. ERP responses in the EEG signal have a poor signal-To-ratio in single-Trial analysis; therefore, the epochs of the target and non-Target trials are averaged over time in order to improve their decoding accuracy. Furthermore, these exogenous potentials can be naturally evoked by just looking at a target symbol. Therefore, the BCI system could generate unintended commands without considering the user's intention. In this study, we approach this dilemma by assuming that a greater effort for the mental task would evoke a stronger positive/negative ERP deflection. Three mental states are defined: passive gazing, active counting, and pitch-imagery. The experiments results showed significantly enhanced ERP patterns and averaged decoding accuracies of 80%, 95.4%, and 95.6%, respectively. The decoding accuracies between both active tasks and the passive task showed an averaged accuracy of 57.5% (gazing vs. counting) and 72.5% (gazing vs. pitch-imagery). Following this result, we proposed a hierarchy classification strategy where the passive or active mental state is decoded in the first stage, and the target stimuli are estimated in the second stage. Our work is the first to propose a system that classifies an intended or unintended brain state by considering the measurable differences of mental effort in the EEG signal so that unintended commands to the system are minimized.