MultiPENG (EEG stream)

Description

MultiPENG was collected at the University of Ottawa (School of Electrical Engineering and Computer Science; Hefeeda is at Simon Fraser University) under the project “Next Generation Cloud Gaming” (NSERC grant ALLRP556311-20). It synchronizes six data streams — EEG, eye tracking, heart rate, gamepad inputs, webcam footage, and gameplay frames — from 39 participants playing two popular titles (FIFA’23 and Street Fighter V) across difficulty levels, with ground-truth engagement labels gathered by the experience-sampling method during natural game pauses (Rashed et al., 2025).

Citation

• Paper DOI: 10.1109/IEEEDATA.2025.3553097
• Dataset DOI: 10.34740/kaggle/ds/6552328
• Reference: Rashed, A., Shirmohammadi, S., & Hefeeda, M. (2025). Descriptor: Multimodal Dataset for Player Engagement Analysis in Video Games (MultiPENG). IEEE Data Descriptions 2, 17–25.

Catalog citation block

Rashed, A., Shirmohammadi, S., & Hefeeda, M. (2025). Descriptor: Multimodal Dataset for Player Engagement Analysis in Video Games (MultiPENG). IEEE Data Descriptions, 2, 17—25. DOI: 10.1109/IEEEDATA.2025.3553097. Dataset (Kaggle): 10.34740/kaggle/ds/6552328, CC BY 4.0.

Download and access

Field	Value
Official source	https://www.kaggle.com/datasets/ammarrashed23/multimodal-player-engagement
Access class	open
Size	18,309,649,997 bytes (~18.3 GB) for the whole bundle, from Kaggle’s own dataset record (totalBytes), version 1 of 2025-01-26. The EEG stream is a small part of it: the bundle also carries eye tracking, heart rate, OpenFace facial features, controller input and 30 fps webcam video per session. No anonymous per-file route was established (see access_steps), so 18.3 GB is the only download size confirmed.
Format	csv
BIDS	false
Mirrors	https://github.com/AmmarRashed/MultimodalEngagement (analysis scripts)

Access steps:

1. §13 item 24, ANSWERED 2026-09-18: NO account is needed. Kaggle’s public API endpoint /api/v1/datasets/download/ammarrashed23/multimodal-player-engagement returns HTTP 302 to a pre-signed storage.googleapis.com URL which serves the bundle; an anonymous range request with no cookies, no API token and no Kaggle credentials returned HTTP 206 and real ZIP bytes (the archive’s PK\x03\x04 magic). So access is ‘open’, not ‘registration’.
2. The website’s own download button does require sign-in: /datasets/ammarrashed23/multimodal-player-engagement/download redirects to /account/login?titleType=dataset-downloads&…&returnUrl=… The two routes disagree; the API route is the one a notebook would use and it is ungated, which is why ‘open’ is recorded.
3. NOT ESTABLISHED: whether a single file can be fetched anonymously. The per-file path /api/v1/datasets/download/// returned 404 for the two Questionnaire/*.csv paths tried, and /api/v1/datasets/list/files// serves a generic HTML page rather than JSON to an unauthenticated caller, so the file list could not be enumerated to test a known-good path. TODO(confirm). Whole-bundle download is the only anonymous route established, and that is 18.3 GB.
4. Access is moot for this site in any case: CC BY-NC 4.0 puts this dataset in §10.7 class C, which ships no snippet and which a notebook never downloads — the lesson links the official access page instead.

Source & access

The dataset is published in IEEE Data Descriptions, vol. 2, pp. 17—25 (2025), paper DOI 10.1109/IEEEDATA.2025.3553097. The data are hosted on Kaggle at kaggle.com/datasets/ammarrashed23/multimodal-player-engagement under dataset DOI 10.34740/kaggle/ds/6552328, released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Analysis scripts are at github.com/AmmarRashed/MultimodalEngagement.

Data structure

Each file is an EmotivPRO CSV export: the canonical header is 163 columns (863 of 900 files; a few have 162 or, for reduced exports, 87). The schema interleaves, per Table III of the descriptor (Rashed et al., 2025, p. 20):

Block	Rate	Columns	Notes
Raw EEG	128 Hz	EEG.AF3 … EEG.AF4 (14) + Counter, Interpolated, RawCq, Battery	µV readings; Interpolated flags dropped-sample fills
Contact quality	128 Hz	CQ.Overall, CQ.<chan>	electrode—scalp contact, 0—4 / 0—100
Signal quality	2 Hz	EQ.Overall, EQ.<chan>	EEG quality, 0—4 / 0—100
Performance metrics	0.1 Hz	31 PM.* cols	Attention, Engagement, Excitement, Stress, Relaxation, Interest (IsActive / Scaled / Raw / Min / Max each)
Facial expression	—	6 FE.* cols	BlinkWink, eye direction, upper/lower face action + power
Band powers	8 Hz	70 POW.* cols (5 bands x 14 ch)	Theta 4—8, Alpha 8—12, BetaL 12—16, BetaH 16—25, Gamma 25—45 Hz

The full Kaggle release is hierarchical (not BIDS): Samples/<pid>/{EEG,EYE,HR,OBS,OpenFace,XBOX}/, plus Questionnaires/ (participants.csv, submissions.csv), splits/fold_[0-6]/ (nested stratified group cross-validation), and a Human Panel Samples/ subset of cropped webcam clips with annotator ratings.

License

• Name: CC-BY-4.0
• Note (verbatim from the directory): CONTESTED between this site’s own catalog and the repository, and the repository is right. §13 item 24 asked only whether a Kaggle account is needed; answering it meant reading the Kaggle record, and the licence there is not the one the catalog holds. Four statements, all read 2026-09-18. (1) The catalog registry records license.name CC-BY-4.0 with license.doi 10.1109/IEEEDATA.2025.3553097. That DOI is the IEEE Data Descriptions descriptor PAPER, and Crossref gives that paper’s own licence as https://creativecommons.org/licenses/by/4.0/legalcode (Unpaywall likewise: is_oa true, oa_status hybrid, licence cc-by). So the recorded CC-BY-4.0 is the ARTICLE’s licence, not the data’s. (2) Kaggle’s own record for the dataset, read from its public metadata endpoint, states licenseName: ‘Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)’. (3) The depositing authors’ own description text on that same Kaggle page says it in their words and gives the reason: ‘This dataset was collected as part of a research study approved by the Research Ethics Board (REB) of the University of Ottawa (Protocol #H-07-23-9439). All participants provided informed consent for their data to be shared for academic and non-commercial research purposes only. In accordance with the ethics approval and participant consent, this dataset is released under a CC BY-NC 4.0 license, which restricts usage to non-commercial applications while requiring appropriate attribution to the original authors.’ It then requires that researchers ‘Use the data solely for academic or non-commercial research purposes’. (4) SILENCES, recorded as silences: the DataCite record for the dataset DOI 10.34740/kaggle/ds/6552328 has an empty rightsList, and the authors’ analysis repository on GitHub carries no LICENSE file. §10.7 already decides this shape of conflict — ‘where the article’s own license differs from the data license (ds-emotions, ds-srm, ds-respect) … the repository’s data license governs’ — so the governing licence is CC BY-NC 4.0 and this dataset is §10.7 class C (non-commercial), not class A. license.name is recorded here as CC-BY-NC-4.0, but the registry wins in build_dataset_pages.build_entry and still states CC-BY-4.0, so license_status: contested is set to close the gate now; snippets derives to ‘no’ by that route today and by the NC name once the registry is corrected. THE FIX BELONGS IN THE REGISTRY, not here: §10.11 item 1 back-fills license.name from prose and item 3 asks for exactly this kind of article-vs-data conflict to be recorded there. The author should set mpeng license.name to CC-BY-NC-4.0 and keep 10.1109/IEEEDATA.2025.3553097 as the paper DOI rather than as a licence DOI. NOT ESTABLISHED: the descriptor paper’s own wording about the data licence could not be read — IEEE Xplore served an HTTP 202 interstitial to an anonymous reader — so whether the paper states CC BY-NC for the data as well is TODO(confirm).
• Snippets on this site: no — the site ships no snippet or precomputed product from this dataset; lessons link the official access page.

Acquisition

Field	Value
Device	14-channel wireless saline-electrode headset (Emotiv EPOC X)
Device class	consumer
Channels	14
Sampling rate (Hz)	128
Online filters	device hardware default (built-in band-pass; the paper does not document an online notch)
Reference	TODO(confirm)
Mains frequency (Hz)	60
Paradigms	video-game, rest-eo, rest-ec
Participants	39
Population	39 adults, mean 24.3 y (30 M / 9 F)
Clinical groups	none
Sessions	TODO(confirm)
Durations	900 annotated micro-game sessions (FIFA’23 mean 91.5 s; Street Fighter V mean 36.7 s; ~694 min total); 15-s EO / 15-s EC baseline at the start of each session (presence in the exports TODO(confirm))

Acquisition

EEG was recorded with an Emotiv EPOC X headset (saline electrodes) running EmotivPRO software, operating at 128 Hz with 14 channels (Rashed et al., 2025, p. 18). The fixed EPOC X montage spans 10—20 positions: AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4. EmotivPRO handles real-time acquisition and computes derived streams in addition to raw EEG. The collection protocol began each session with eye-tracker calibration and a 15-s eyes-open / 15-s eyes-closed EEG baseline; electrode contact was verified with saline before recording. Online filtering is the EPOC X hardware default (a built-in Sinc bandpass; the paper does not separately document an online notch).

Participants

39 participants — 30 male, 9 female, mean age 24.3 years (Rashed et al., 2025, p. 18). Participants ranged from people with no prior gaming experience to casual players and experts in the selected titles. Written consent was obtained and the protocol was approved by the University of Ottawa Office of Research Ethics and Integrity (file H-07-23-9439). The sample is non-clinical with one documented exception: participant ID 559 reported a prior ADHD diagnosis and was taking stimulant medication during the experiment (see caveats).

Tasks / conditions

Participants played two contrasting commercial titles — FIFA’23 (sports; 6 difficulty levels: Beginner, Amateur, Semi-Pro, Professional, World Class, Legendary) and Street Fighter V / SFV (fighting; 8 numeric difficulty levels, 1 = easiest to 8 = hardest) (Rashed et al., 2025, p. 19). The corpus comprises 900 annotated micro-game sessions (Table I): per session durations averaged 91.5 s (SD 50.3) for FIFA and 36.7 s (SD 9.8) for SFV, totalling ~694 minutes. After each natural break point (FIFA: after goals, half-time, post-match, with a 20-s minimum between surveys; SFV: between rounds), participants answered a four-dimension survey on 5-point Likert scales (Table II): Engagement (Very Bored 0 -> Very Engaged 4), Interest (Strongly Disliked 0 -> Strongly Liked 4), Stress (Very Relaxed 0 -> Very Stressed 4), and Excitement (Not Excited 0 -> Extremely Excited 4). These four scores are encoded directly in each CSV filename.

Used on this site

L7.5

Loader

Fetches only the stated subset. Recorded loader: Not loaded by this site. §10.7 class C: the lesson links the Kaggle page and downloads nothing..

# Kaggle host; TODO(confirm) whether an account is required for download.
kaggle datasets download -d ammarrashed23/multimodal-player-engagement --unzip

Caveats worth knowing

• LICENCE: CC BY-NC 4.0 per the repository and per the depositing authors’ own words, not the CC BY 4.0 the catalog registry records — the registry appears to have captured the descriptor article’s licence instead of the data’s. No asset derived from this dataset ships, and no notebook downloads it, until the author corrects the registry (§10.11 items 1 and 3). See license.note for all four statements.
• Consent-limited, not merely licence-limited: the authors state that participants consented to sharing ‘for academic and non-commercial research purposes only’ under University of Ottawa REB protocol #H-07-23-9439. The NC term is downstream of the consent, so it is not the kind of restriction a licence decision can trade away.
• 18.3 GB as a single bundle with no established per-file anonymous route, for a dataset whose EEG is a small fraction of the payload. Even with a permissive licence this would be a poor fit for a notebook on a small machine.
• Only ~50% of samples meet the authors’ quality criterion (vendor-computed EQ.OVERALL ≥ 75%) during intense play.
• The CSV interleaves 14 EEG columns with contact-quality, “performance-metric” and band-power columns that are vendor-derived, not raw signal.
• Nominal 128 Hz output but ≈ 43 Hz hardware bandwidth and built-in 50/60 Hz notches (the descriptor documents only the built-in band-pass; notch and ceiling are inferred from the device specification).

Notable caveats

• Not a clean EEG matrix. Each CSV row interleaves 14 EEG channels with contact-quality, signal-quality, performance-metric, facial-expression, and precomputed band-power (POW) columns; loaders must select the EEG.* columns explicitly.
• EEG quality drops during intense play. Only ~50% of samples have an aggregate EEG quality (EQ.OVERALL) >= 75% (Rashed et al., 2025, Fig. 3); the authors attribute this to sudden head/body movement during highly engaging moments. Quality-threshold filtering on CQ/EQ is recommended before spectral analysis.
• Missing / atypical sessions. Controller input was not captured for 10 participants (IDs 120, 166, 462, 539, 623, 703, 754, 507, 514, 744); eye-tracking was incomplete for ID 407; six participants (IDs 872, 850, 568, 533, 297, 183) were recorded under different lighting (affects webcam/OpenFace, not EEG); and ID 559 (ADHD, on stimulants) shows a diminished EEG quality score consistent with stimulant effects on EEG. None of these break the EEG CSVs but they matter for cross-modal work.
• Mains / line noise. Collected in Ottawa, Canada, so mains is 60 Hz.
• Bandwidth ceiling. The highest documented POW band is Gamma 25—45 Hz.
• Two titles only. Engagement is sampled across just FIFA’23 and SFV, a deliberate genre contrast but a coverage limit the authors flag explicitly.

Related datasets

• ArEEG_Words
• EEG for natural-image recognition (VEP)
• EEGEmotions-27
• PhySF, Physiological Sense of Flow
• ERP CORE (Compendium of Open Resources and Experiments)

# Loader recorded in the catalog: Not loaded by this site. §10.7 class C: the lesson links the Kaggle page and downloads nothing.
# TODO(confirm) the exact call and the subset to fetch. Official source: https://www.kaggle.com/datasets/ammarrashed23/multimodal-player-engagement