The objective of the RoboCamp project is to enhance the attention capabilities of children diagnosed with ADHD by employing feedback with the virtual robotic agent Flobi. To expand the current gaze-based detection of attention drops, this contribution investigates the feasibility of integrating EEG based attention detection into the RoboCamp system. An analysis of EEG recordings from 64 channels during supervised tasks with 67 individuals, including both neurotypical and participants with ADHD, identified significant differences between hits and misses among participants, indicating attentional variations. Machine learning models were trained on EEG time sequences derived from experiment trigger signatures, exploring features such as time series, engagement index, and frequency band powers. By applying the features in a classification task, it was found that the EI is not effective for detecting inattention in the ADHD group (F1-score: 57.3%) while it yields better results in the NT group (F1-score: 67.3%). The time series and band power features showed classification capabilities comparable to those of the neurotypical group. The transfer within the neurotypical group to newly recorded RoboCamp tasks was successful. Interestingly, the classifier yielded an F1-score of 92.8% for NT participants on a task involving math exercises, indicating that the task itself may play an important role in whether or not inattention can be automatically detected.