Nonlinear EEG Analysis for Distinguishing Mind Wandering and Focused Attention: A Machine Learning Approach

Kheiri, Farshad; Sundaram, Shyam; Bragin, Anatol · 2024 · Crossref

DOI: 10.1101/2024.10.18.618974

archive: archived pipeline: cataloged verified

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

Summary

This study investigates the use of nonlinear electroencephalogram (EEG) analysis to distinguish between mind wandering (MW) and focused attention (FA) states. While linear EEG measures provide limited insight into the brain's dynamic complexity, nonlinear techniques offer a framework for capturing the chaotic and irregular neural patterns underlying distinct cognitive states. The research aims to evaluate the efficacy of various nonlinear features in classifying these mental states using machine learning, addressing the need for more advanced methods to characterize the brain's nondeterministic nature during introspective versus focused tasks. The study utilized EEG data from a single participant across 21 recording sessions. Each session involved a 6-minute recording with eyes closed, beginning with 5 minutes of MW, where the participant recalled past experiences, followed by a 50 dB wooden block click that signaled a transition to 3 minutes of FA, involving mindful breathing. EEG signals were recorded using a 32-channel system at 250 Hz. Data were cleaned using EEGLAB, including artifact removal and independent component analysis. The analysis focused on 30-second windows surrounding the transition cue, yielding 42 total segments (21 MW, 21 FA). Seven nonlinear features were extracted: Global Field Power, Global Frequency, Global Complexity, Mean Power, Standard Deviation of Power, Mean Frequency, and Standard Deviation of Frequency. These features were calculated across segment durations of 2, 3, 5, 6, 10, and 15 seconds. Machine learning models, including gradient boosting trees, random forest, support vector machines, and logistic regression, were trained to classify the states, with 10-fold cross-validation employed to prevent overfitting. The gradient boosting trees algorithm demonstrated superior performance among the tested models. The highest classification accuracy of 75% was achieved using 5-second segments, outperforming longer durations such as 15 seconds (63.90%) and shorter ones like 2 seconds (55.56%). Feature importance analysis revealed that frequency-related metrics were the most critical discriminators. Mean frequency was the most influential feature, followed by global frequency and the standard deviation of frequency. This indicates that dynamic changes in brainwave frequencies are key markers for distinguishing MW from FA, rather than power or complexity measures alone. The findings underscore the potential of nonlinear EEG analysis in revealing the complex neural dynamics associated with different cognitive states. The superior performance of shorter time segments suggests that temporal variations in cognitive states are best captured in brief windows, highlighting the importance of temporal resolution in EEG classification. The study concludes that frequency-based nonlinear features are robust indicators for differentiating MW and FA. Future research should explore temporal dynamics using recurrent neural networks, develop personalized models to account for individual variability, and investigate the neural mechanisms driving these frequency changes to advance applications in cognitive enhancement and mental health interventions.

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 qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

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

• neuro workload indices