Active Task Engagement Enhances Auditory Brain–Behaviour Prediction From Single‐Trial EEG Compared With Passive Listening

Auditory neural processing during active task engagement and passive listening reflects distinct task contexts with potentially different behavioural relevance. While both contexts elicit deviance-related responses, it remains unclear, which yields neural measures that more reliably predict individual differences in behavioural performance. To address this question, we employed a multi-feature auditory paradigm in which frequency, duration, and intensity deviants were presented under passive (no response required) and active (explicit detection required) conditions. EEG was recorded from 47 participants; passive listening was characterized by a prominent mismatch negativity (MMN), whereas active discrimination was characterized by an additional P3b component. Beyond conventional ERP measures, we quantified individual-level neural discriminability using EEGNet, a neural-network-based classifier, by classifying deviant versus standard single-trial epochs and deriving cross-validated decoding accuracy. Behavioural performance was quantified using an efficiency score (ES) that integrates hit rate and reaction time. Participants were stratified into high- and low-performance groups based on a median split of ES. Results showed that the expected MMN during passive listening and the P3b during active discrimination were elicited, as confirmed by spatiotemporal cluster-based permutation analysis. Furthermore, decoding accuracy derived from the active discrimination condition robustly separated high- and low-performance groups (Group × Task: F = 29.62, p < 0.001) and predicted behavioural efficiency across individuals (r = 0.53, p < 0.01). In contrast, passive-listening decoding showed reduced overall discriminability and minimal group separation. Together, these findings indicate that task engagement amplifies the behavioural relevance of single-trial neural discriminability, enabling stronger auditory brain-behaviour prediction than passive listening.