Mobile EEG Datasets: Real-World Neurodata

Updated 27 January 2026

Mobile EEG datasets are systematically acquired collections of EEG recordings captured outside traditional labs using wearable, portable sensors.
They integrate multimodal data such as IMU, EOG, and behavioral annotations to enable robust artifact correction and detailed signal analysis.
Their applications span mobile BCIs, clinical monitoring, and real-world neuroengineering, driving open-science research and algorithm benchmarking.

Mobile EEG datasets are systematically acquired, annotated collections of electroencephalographic recordings obtained outside traditional laboratory or clinical settings, often leveraging wearable, portable, or consumer-grade EEG hardware. These datasets are central to research domains spanning mobile brain-computer interfaces (BCI), cognitive monitoring, assistive neurotechnology, translational neuroscience, and the development of next-generation algorithms for artifact rejection and physiological event detection. The transition from stationary, high-density to mobile, low-to-moderate channel-count or dry-electrode systems has spurred numerous open datasets and methodological innovations, with implications for accessibility, scalability, and robustness in real-world environments.

1. Representative Mobile EEG Datasets

Mobile EEG datasets vary across clinical populations, use cases, sensor modalities, and experimental protocols. The following exemplars span key application domains:

Dataset Name	Hardware/Channels	Scenario/Population
Mobile BCI dataset	32 scalp, 14 ear, 4 EOG, 27 IMU	24 healthy adults, treadmill BCI (ERP/SSVEP)
MODMA (Mobile EEG Arm)	3 frontal electrodes (Fp1, Fpz, Fp2)	55 adults (depressed & controls), eyes-closed rest
NEUROSKY–EPI	1 frontal (Fp1, dry-electrode)	25 epilepsy pts., clinic, rest + awake
Neural Tracking AVEEG	44 scalp, 20 cEEGrid	24 normal-hearing, AV attention & conversation
Consumer-grade EEG–ET	5 (TP9, TP10, AF7, AF8, Fpz ref)	113 adults, 4 eye-movement tasks, webcam-tracked

Each dataset employs distinct sensor configurations, ranging from ultra-compact 1–3 channel systems (enabling affordable, pervasive monitoring) to moderate-density arrays (for detailed source mapping or auditory processing) (Lee et al., 2021, Cai et al., 2020, Tabib et al., 22 Oct 2025, Wilroth et al., 21 Jan 2026, Afonso et al., 18 Mar 2025).

2. Signal Acquisition: Modalities, Hardware, and Annotation

Mobile EEG datasets leverage a diversity of recording devices:

Scalp and ear-EEG: E.g., the Mobile BCI dataset uses 32 scalp electrodes (Ag/AgCl, 10–20 system), two cEEGrid ear arrays (left: 8, right: 6, plus grounds), and synchronized IMUs for precise motion labeling (Lee et al., 2021).
Ultra-wearable/Consumer-grade: MODMA employs UAIS LAB BIAS V2.0 (3 frontal electrodes, 24-bit), NEUROSKY–EPI employs MindWave Mobile 2 (single Fp1, dry, ∼$100 device), and the EEG–ET dataset uses Muse S2 Headband (5 dry electrodes) (Cai et al., 2020, Tabib et al., 22 Oct 2025, Afonso et al., 18 Mar 2025).
Multimodal augmentation: Many datasets integrate EOG (to capture ocular artifacts), audio, behavioral response streams, and IMUs (kinematics).

Annotation structures are standardized when possible:

EEG-BIDS/BrainVision Core (Mobile BCI) and hierarchical subject/session/task directories encode paradigm and speed conditions (Lee et al., 2021).
Metadata include clinical diagnosis (e.g., DSM, PHQ-9 in MODMA), session parameters, and behavioral assessments (Cai et al., 2020, Tabib et al., 22 Oct 2025).

3. Experimental Paradigms and Protocols

Mobile EEG datasets sample a broad spectrum of experimental manipulations:

Movement and BCI: E.g., the Mobile BCI dataset records event-related potentials (ERP) and steady-state visual evoked potentials (SSVEP) with participants standing, walking, and running at precisely set treadmill velocities (0, 0.8, 1.6, and 2.0 m/s). ERP stimuli: 500 ms character flashes (“OOO” vs. “XXX”), randomized intervals; SSVEP: three square flickers (5.45, 8.57, 12 Hz) (Lee et al., 2021).
Clinical States: MODMA’s mobile EEG arm focuses on 90 s eyes-closed resting state recordings in a quiet clinical environment, emphasizing frontal θ/α/β, coherence, and asymmetry indices for mood monitoring (Cai et al., 2020). NEUROSKY–EPI records single-channel, 60 s rest and 60 s light-cognitive-task (eyes open) in epilepsy patients (Tabib et al., 22 Oct 2025).
Attention and Auditory Environments: The AVEEG dataset presents sustained, switching, and conversational attention tasks in two-talker environments, using synchronized audiovisual stimuli, detailed event marking, and behavioral comprehension probes (Wilroth et al., 21 Jan 2026).
Eye-Movement Tracking: EEG–ET aligns gaze trajectories (from webcam-based eye-tracker) with 5-channel EEG under smooth pursuit and saccade paradigms of varying complexity (Afonso et al., 18 Mar 2025).

4. Preprocessing Pipelines and Signal Quality Metrics

Artifact-prone mobile EEG necessitates robust, transparent preprocessing, commonly documented and often provided as shared code:

High-pass/Notch/Bandpass Filtering: Example: 0.5 Hz (Butterworth, 5th order) for mobile BCI; 1–45 Hz FIR for MODMA; 0.1–40 Hz zero-phase FIR in AVEEG (Lee et al., 2021, Cai et al., 2020, Wilroth et al., 21 Jan 2026).
Artifact Correction: Adaptive regression for EOG (flt_eog in BBCI/BCILAB), iterative ANC for eye-blinks in MODMA, no ICA for some consumer datasets (Muse/NEUROSKY), extended Infomax ICA for AVEEG (scalp/cEEGrid) (Lee et al., 2021, Cai et al., 2020, Wilroth et al., 21 Jan 2026).
Channel Rejection/Interpolation: Statistical z-score thresholds (SD z > 4) and spherical interpolation, with average channels interpolated per session (e.g., 2.4 ± 1.9 for scalp, 1.4 ± 1.2 for ear in Mobile BCI) (Lee et al., 2021).
Re-referencing: Common average (scalp), ear array reference (cEEGrid), explicit per device documentation.

Quality and noise benchmarks are dataset-specific:

ERP SNR at Pz: $SNR_{ERP} = RMS\{\text{P300}\}/RMS\{\text{baseline}(-200–0~\text{ms})\}$ .
SSVEP SNR at Oz: $SNR_{SSVEP} = P(f_{target}) / \text{avg}[P(\text{neighbors})]$ .
Band-limited power changes: Clustering-based permutation shows $\delta$ -band (0.5–3.5 Hz) increases with gait, and both ERP AUC and SSVEP scoring decline with speed (ERP AUC: 0.90 → 0.67, SSVEP accuracy: 88.7% → 80.7%) (Lee et al., 2021).

Consumer-grade pipelines (Muse S2 Headband, NEUROSKY MindWave) are tailored for missing-value imputation (Kalman/SARIMA), basic artifact exclusion, and time–frequency representation (Welch method, CNN/SVM features) (Cai et al., 2020, Tabib et al., 22 Oct 2025, Afonso et al., 18 Mar 2025).

5. Feature Extraction and Analysis Frameworks

Feature extraction in mobile EEG datasets spans spectral, temporal, spatial, and machine learning domains:

Spectral: Welch’s method for PSD; bandpower integration for θ, α, β; real-time asymmetry gradients, e.g., $A = \log(P_\alpha(Fp2)) - \log(P_\alpha(Fp1))$ ; intra-electrode coherence $C_{xy}(f)$ (Cai et al., 2020).
Temporal: Epoch-based ERP/SSVEP quantification, event-aligned averaging, signal-to-noise analysis as above.
Spatial: Electrode correlation (e.g., $r_{ij}$ in walking vs. running), TRF-based source modeling (Lee et al., 2021, Wilroth et al., 21 Jan 2026).
Data-driven/ML: Logistic regression, cross-validated SVM/1D-CNN (MODMA), transfer learning (EEGNet–EmbedCluster pipeline), unsupervised clustering (K-means, GMM), autoencoder embeddings (NEUROSKY–EPI) (Cai et al., 2020, Tabib et al., 22 Oct 2025).
TRF Modeling (AVEEG): Forward and backward modeling equations for neural tracking:

$\hat{y}_i(k) = \sum_{l=l_1}^{l_2} h(l,i)\, x(k-l),\quad \hat{x}(k) = \sum_{i=1}^{n_{ch}}\sum_{l=l_1}^{l_2} g(l,i)\, y_i(k+l)$

with lag optimization and Hamming-windowed filter basis (Wilroth et al., 21 Jan 2026).

6. Applications, Limitations, and Open-Science Access

Mobile EEG datasets enable:

Algorithm benchmarking: Realistic BCI decoding under movement, validation of artifact-rejection strategies with synchronized IMU/eye-tracking ground truth, functional network inferences, and event-detection paradigms (Lee et al., 2021, Afonso et al., 18 Mar 2025).
Clinical and ambulatory monitoring: Mobile mental state tracking (MODMA, NEUROSKY–EPI) in resource-limited or community settings; patient stratification for epilepsy care (Cai et al., 2020, Tabib et al., 22 Oct 2025).
Real-world neuroscience: Ecologically valid studies of attention switching, natural conversation, EEG-based interaction, and human–machine co-adaptation (Wilroth et al., 21 Jan 2026, Afonso et al., 18 Mar 2025).

Key limitations include:

Motion-induced contamination: High-frequency movement yields low-frequency EEG artifacts, mandating robust preprocessing (≥0.5 Hz filtering, regression-based correction, ICA is less common in ultra-wearable devices) (Lee et al., 2021).
Sparse spatial coverage: Single- or three-channel headsets limit source resolution and SNR for occipital features; advanced denoising and multimodal fusion are required for complex decoding (e.g., SSVEP, mood estimation) (Cai et al., 2020, Tabib et al., 22 Oct 2025).
Instrumentation artifacts and missing data: Consumer platforms may suffer dropouts, channel failure, or poor skin contact. “Missing_data” files and manual flagging mitigate, but do not eliminate, the issue (Lee et al., 2021, Afonso et al., 18 Mar 2025).
Ambiguity in state or context labels: Non-laboratory acquisition may lack behavioral ground truth for intended tasks, challenging benchmarking and generalizability (Afonso et al., 18 Mar 2025).

Access to raw and processed data, metadata, and code is typically provided via open repositories (e.g., OSF, Zenodo, GitHub) with CC-BY or EULA licensure; datasets often include cross-modal alignment and scripts for data loading and analysis (Lee et al., 2021, Wilroth et al., 21 Jan 2026, Afonso et al., 18 Mar 2025).

7. Future Directions and Research Significance

Mobile EEG datasets have catalyzed research in methods robust to real-world noise and sensor limitations, democratized ambulatory neurophysiology, and paved the way for scalable, context-aware BCI and monitoring systems. A plausible implication is that the increasing fusion of EEG with IMU, eye tracking, and contextual metadata (e.g., self-report, clinical annotation) will further enable signal separation, neuro–behavioral inference, and closed-loop interventions under free-living conditions.

Challenges remain in standardization, interoperability, and generalization of findings across device classes, populations, and application domains. Ongoing open-science practices, diverse cohort recruitment, and multi-institutional benchmarking are primary drivers in establishing robust, inclusive, and actionable mobile EEG research (Lee et al., 2021, Cai et al., 2020, Tabib et al., 22 Oct 2025, Wilroth et al., 21 Jan 2026, Afonso et al., 18 Mar 2025).