Distinguishing vigilance decrement and low task demands from mind‐wandering: A machine learning analysis of EEG

Jin, Christina Yi; Borst, Jelmer P.; van Vugt, Marieke K. · 2020 · Crossref

DOI: 10.1111/ejn.14863

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

**Research Question and Motivation** This study investigates whether mind-wandering is a distinct mental state or merely a consequence of low vigilance and low task demands. Previous research suggests that mind-wandering, vigilance decrement, and low task demands share behavioral consequences (e.g., increased errors) and neural correlates (e.g., default mode network activation, reduced P3b, increased alpha power). The authors hypothesized that if these states are equivalent, machine learning classifiers trained to detect low vigilance or low task demands should successfully predict mind-wandering. Conversely, if mind-wandering is an independent phenomenon, such cross-prediction should fail. **Methods** Thirty participants performed two tasks while EEG was recorded: a visual search task manipulating task demands (high-demand counting vs. low-demand passive viewing) and a Sustained Attention to Response Task (SART) used to measure vigilance over time. Participants provided self-reports of their mental state (mind-wandering vs. on-task) via intermittent thought probes. The researchers extracted alpha-band (8.5–12 Hz) power from independent components (ICs) of the EEG data. Support Vector Machine (SVM) classifiers were trained on the visual search task data using three labeling schemes: (1) task demands (counting vs. non-counting), (2) vigilance (first half vs. second half of the task), and (3) self-reported mental state. These classifiers were then tested on SART data, which was labeled exclusively by self-reports. Additionally, dipole fitting was used to source-localize the neural generators of the most predictive features. **Findings** Behavioral data confirmed that high task demands and high vigilance were associated with faster response times and higher accuracy. Critically, the machine learning results showed that neither the vigilance classifier nor the task demands classifier could predict mind-wandering in the SART above chance level. In contrast, the classifier trained on self-reported mind-wandering successfully predicted mind-wandering in the SART. Source localization revealed that the neural structures associated with the predictive features for mind-wandering were distinct from those associated with low vigilance or low task demands. **Significance** The findings demonstrate that mind-wandering is a qualitatively different mental state from low vigilance or low task demands, despite their overlapping behavioral and oscillatory signatures. The study highlights the utility of machine learning classifiers in distinguishing subtle neural patterns that standard statistical averages might obscure. By combining classifier analysis with source localization, the research provides a methodological framework for identifying the specific neural substrates underlying distinct attentional states, advancing the understanding of the independence of mind-wandering from general attentional lapses.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices
• sustained attention vigilance
• vigilance
• mind wandering
• drowsiness detection algorithms

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data