EEG Spectral Analysis for Inattention Detection in Academic Domain

The increasing rate of attention deficit among students, attributed to social media, has far-reaching consequences on their academic performance. Inattention or lack of attention is a state of absent-mindedness or not paying enough attention to the details. A rich body of literature suggests that it is highly associated with underachievement in the academic context. In particular, students who have been clinically diagnosed with Attention Deficit Hyperactivity Disorder(ADHD), a neuro-developmental disorder characterized by inattention, hyperactivity, and impulsivity symptoms are more at risk of under-performance and retention, with some leaving school without a terminal degree. In addition to the academic domain, inattention generally impacts the quality of life and future occupations. Research suggests that ADHD is also attributed to societal and unemployment excess costs as well as productivity loss and healthcare expenses which are estimated to be over $14K per adult in the United States (US). Although more than eight million adults were identified with ADHD in the US by 2018, not all inattention cases are associated with ADHD. Inattention could be caused by several other factors such as stress or anxiety and its early detection and timely intervention is critical, especially in the academic domain. There are existing studies that analyze brain signals by Electroencephalogram (EEG) scans to identify individuals who have ADHD. In this study, we developed a Machine learning (ML) pipeline model that is trained on both ADHD and inattention data to determine if a person is having an attention problem. In the first phase of developing the model, we trained a few different classifiers on a 19-channel public EEG dataset of 60 ADHD and 61 non-ADHD participants. Data analysis showed K-Nearest Neighbor (KNN) classifier outperformed other classifiers with an accuracy of 89%. While many of the existing papers focus on ADHD data, in this work we expand our model to analyze the attention deficiency data as well. To train the model we used an attention dataset which is a collection of 34 recordings of 14-channel EEG that scan the attention states of five young adults. We further implemented Independent Component Analysis (ICA) to reduce noise and dimensions and tried different classifiers to improve the accuracy of the model. We achieved the highest accuracy of 98% with ensemble model classifiers using the improved ML pipeline. To evaluate the proposed model, we conducted a study to record the EEG of 15 young adults while taking a visual reasoning attention test for the duration of 8 minutes. We compared the model output (i.e. binary attention label) and their average test score. Findings showed consistent results between the proposed models' prediction and the visual attention test. this indicates there is a potential to use that test as an alternative to the EEG recording to assess students' attention levels in the academic domain. The goal of this research is not for any diagnosis purposes but merely as a tool to help in the early identification of inattention for timely help and interventions among students.