Applying bayesian networks to brain data

Understanding how different parts of the brain communicate is central to studying human thought and behaviour. Electroencephalography (EEG), which records electrical activity from the scalp, offers a low-cost way to measure these interactions. This project explores the use of Bayesian networks (BNs), a mathematical framework that models how variables influence one another, to capture patterns of brain connectivity. While BNs are widely used in biological applications in general, they have rarely been applied to EEG data. Recordings from 64 EEG sensors were taken from participants while they listened to an audiobook. They did this three times, each under a different condition: no task (listening casually), semantic task (listening carefully), and auditory task (focusing on speaker accents). For each participant and task, a BN was derived from their EEG recording, which describes how neural activity flows between brain regions. This allows us to visualise connectivity. Machine learning models were then used to test whether the BN-derived patterns could distinguish between the three mental states. The results showed that connectivity patterns identified by BNs preserved meaningful differences between the brain states, and classification models could reliably tell them apart. This suggests that BNs can provide interpretable and powerful descriptors of brain activity. This research demonstrates the potential of BNs as a new tool for EEG analysis. Beyond advancing neuroscience methods, it may support applications in education, clinical diagnostics, and brain–computer interfaces. The next step is to refine these models and test them on larger, more diverse datasets to assess their generalisability.