A Causal Machine Learning Approach for Investigating Learners’ Mental States through Electroencephalography (EEG)

In recent years, the non-invasive method of Electroencephalography (EEG) has gained widespread acceptance as a means of using computers to comprehend the intricate psychological behaviour of the human mind. The study of brain-computer interface via EEG offers additional compelling insights into human psychology. Nevertheless, it is difficult to employ EEG in brain contact successfully due to the extremely low band frequency and great sensitivity of the signals to noise and muscle activity, which leads to complex jobs in computation and machine learning. To increase the performance of brain-computer interaction, this work presents a novel combination of statistical causal inference with the machine learning approach for two different cases. The first case is a publicly available dataset that classifies a perplexed learners’ responses disclosing confusion level of a learner to stimuli of Massive Open Online Courses (MOOC), whereas the second case classifies a stressed extent of a learner while watching multiple stimuli videos. The propensity score matching is used to calculate weights of balance of non-EEG variables and incorporated them into the causal model with brain wave feature engineering techniques used for machine learning. SVM, Random Forest, Gradient Boosting for Additive Model, Bagged CART Model and Bayesian Generalized Linear Model algorithms are used to validate the proposed method. Results have shown that LightGBM outperforms with 98.44% accuracy over other models with SVM(≈96%), Gradient Boosting for Additive Model (≈91%), Bagged CART Model(≈97%) and Bayesian Generalized Linear Model(≈58%). The proposed approach significantly improves classification accuracy by incorporating causal inference through Propensity Score Matching (PSM), which directs confounding factors typically ignored in traditional methods.