Researchers at Fırat University and University of Southern Queensland introduce OTPat, an explainable feature engineering pipeline that leverages order transition patterns, CWINCA feature selection, and tkNN classification to achieve over 95% accuracy in EEG and ECG signal classification focused on stress, ALS, and mental health conditions.

Key points

• OTPat uses ordering transformers and transition tables to extract spatial-temporal features from EEG/ECG signals.
• CWINCA applies normalized NCA weights and cumulative thresholds to auto-select the most informative features.
• tkNN generates 90 parametric kNN outcomes and 88 iterative-voted results, choosing the highest-accuracy classification.
• Framework achieves 99.07% on EEG stress, 95.74% on EEG ALS, and 100% on ECG mental health datasets.
• DLob and Cardioish symbolic languages produce interpretable connectome diagrams and entropy metrics.

Why it matters: This framework offers a computationally efficient alternative to deep learning for biomedical signal classification, achieving high accuracy while generating interpretable connectome diagrams. Its explainable outputs and linear-time complexity can facilitate broader clinical adoption in diagnosing stress-related, neurological, and mental health disorders.