Multimodal Sensing Layer

EEG | ECG | EMG | EDA | PPG | Eye Metrics

Contextual Inputs (task load, duration)

Edge Processing Layer

• Signal denoising & normalization

• Segmentation & windowing

• Feature extraction

(EEG spectral power, HRV, EMG amplitude, blink rate, skin conductance)

[On-device preprocessing | Low power]

Adaptive Learning Layer

Lightweight Models:

• Quantized CNN / 1-D CNN

• Random Forest / SVM

• Tree-Augmented Naïve Bayes (TAN)

Optional: Federated Weight Synchronization

(No raw data sharing)

[On-device inference | Energy-aware]

Explainability Layer

• Feature importance scores

• Conditional dependency graphs

• Confidence and uncertainty estimates

[Human-interpretable outputs]

Decision & Feedback Layer

• Fatigue level classification

• Risk alerts & thresholds

• Adaptive workload or rest recommendations

Feedback loop to Adaptive Learning Layer

Model update / personalization