The code in this repository was developed for my Bachelor's thesis, an investigation titled "Classifying Attention States: Contrasting the Performance of Machine and Deep Learning models", that sought to investigate the possible difference in performance both types of models display. The models chosen were: Random Forest and Bidirectional Long Short Term Memory (BiLSTM) network and the abstract is attached below. Code within this repository includes: Code for Random Forest, Code for BiLTSM, Code for generating Confusion Matrices for both as well as the resulting images generated.
Abstract The ability to accurately classify attention states based on EEG signals holds significant promise for brain-computer interface (BCI) systems and cognitive monitoring applications. Although previous studies have been able to classify attention or distraction, further exploration of more nuanced attention states is lacking. In this paper, three attention states (focused, unfocused, and drowsy) are classified by both Machine and Deep Learning models, in particular Random Forest and BiLSTM. The results show that BiLSTM consistently outperforms RF in terms of accuracy (90% v 73%) and the F1 score (0.87 v 0.59), particularly in distinguishing subtle attention changes.