MOCAS Dataset
"MOCAS: A Multimodal Dataset for Objective Cognitive Workload Assessment on Simultaneous Tasks"
Overview
MOCAS is dedicated to cognitive workload assessment in realistic monitoring tasks. Unlike most datasets based on abstract stimuli (N-Back, Stroop), MOCAS uses closed-circuit television (CCTV) monitoring as its primary task - a realistic operational surveillance scenario with high ecological validity.
The use of a realistic monitoring task makes MOCAS substantially more relevant to real-world deployments than lab-only datasets.
Study Design
- 21 participants
- Participants monitored multiple CCTV feeds simultaneously, managing their attention across streams to detect events and respond to queries.
- Two off-the-shelf wearable sensors complemented by a standard webcam.
Cognitive Task: CCTV Monitoring
Participants monitored between 1 and 6 simultaneous CCTV feeds (varying across conditions), creating controlled variation in cognitive load through multi-stream attention demands. The monitoring task involves:
- Detecting events occurring in any of the monitored feeds.
- Maintaining awareness of ongoing activity across all streams.
- Responding to queries about events that occurred.
This is structurally analogous to many real-world monitoring roles, where operators must maintain awareness across multiple concurrent streams and detect anomalies as they develop.
Cognitive Labels
| Instrument | Description | Administration Timing |
|---|---|---|
| NASA-TLX | Six-dimension composite workload score | Post-condition |
| ISA (Instantaneous Self Assessment) | Single 5-point real-time workload rating | During task |
| SAM (Self-Assessment Manikin) | 9-point arousal + 9-point valence scales | Post-condition |
Additionally collected:
- Demographic questionnaire - age, gender, experience with surveillance tasks.
- Big Five Factor personality questionnaire - openness, conscientiousness, extraversion, agreeableness, neuroticism.
The Big Five personality data enables investigation of individual differences as conditioning variables: is high neuroticism associated with different physiological workload signatures? Does extraversion correlate with different baseline EEG patterns? These questions are relevant to personalisation research.
Technical validation confirmed that the task design successfully elicited the target CWL levels across conditions.
Sensor Specifications
Emotiv Insight (EEG)
| Property | Value |
|---|---|
| Channels | 5 |
| Electrode positions | AF3, AF4, T7, T8, Pz |
| Technology | Dry electrodes |
The Emotiv Insight's electrode layout (frontal AF3/AF4, temporal T7/T8, parietal Pz) covers: - Prefrontal cognitive control (AF3, AF4 near PFC). - Temporal working memory and emotion (T7, T8 near hippocampus and amygdala). - Parietal integration (Pz in the DMN region).
Webcam (Behavioural)
A standard webcam captures:
- Facial expressions and micro-expressions.
- Blink rate and duration.
- Gaze direction (approximate).
- Head position and movement.
Behavioural features derived from webcam data are complementary to the physiological EEG and PPG signals, enabling multimodal fusion experiments that include computer-vision-derived features.
Why MOCAS is Valuable
| Property | Value for Brain FM |
|---|---|
| Ecological validity | CCTV monitoring elicits realistic multi-stream vigilance demands |
| ISA during task | Real-time workload labels enable time-aligned workload prediction evaluation |
| SAM (arousal + valence) | Enables joint workload + affect modelling |
| Big Five personality | Enables individual difference studies; potential conditioning for personalisation |
| Webcam modality | Tests fusion of physiological + behavioural signals |
Relevance to Brain FM
MOCAS is used for:
- EEG fine-tuning - Emotiv Insight EEG with NASA-TLX / ISA labels; tests performance on very low channel count (5 ch) with frontal-temporal placement.
- Multimodal fusion - EEG + webcam-derived behavioural features.
- Individual differences research - Big Five personality as a covariate in personalisation experiments.
- Cross-dataset generalisation - training on UNIVERSE or WAUC and evaluating on MOCAS tests cross-dataset, cross-hardware transfer.