Investigation on Cost-Sensitivity in EEG-Based Confusion Emotion Recognition Systems via Ensemble Learning

Confusion is an important academic emotion that usually arises when a student cannot integrate new information with existing knowledge. Identifying confused students for an educator is a very challenging task as students may not express their confusion state. Recently, EEG – based confusion emotion recognition systems had been developed using machine learning as an aid for educators to help identify their confused learners easily. But the problem with those studies were that the systems were developed for experimental purposes and not for real world usage. Due to this, educators are not able to experience the benefits of such systems. In this study, we explored the possibility of developing a system fitting for a practical application using the concepts in cost sensitive machine learning. We developed two heterogenous machine learning models (Serial and Parallel combination) with the aim of achieving a tradeoff between prediction performance and cost sensitivity in terms of reduced misclassification cost. Results showed that the designed serial combination ensemble model achieved the best tradeoff between precision (highest value of 62.93%) and misclassification cost (lowest value of 1811.75) which could not be achieved by independent classifiers.