Supplementary Materials for "Interpretable machine learning for cardiogram-based biometrics"

Q: What is the clinical evidence from this study?

Study design: Cross-Sectional. Population: Healthy (n=202). Intervention: Anger induction vs. Baseline (relaxation). Primary outcome: Differences in cardiogram-based features between baseline and anger states.

Q: What does this research mean for the field?

The supplementary materials enhance the use of interpretable machine learning for cardiogram-based biometrics. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Ilija Tanasković; Ljiljana B. Lazarević; Goran Knežević; Nikola Milosavljević; Olga Dubljević; Bojana Bjegojević; Nadica Miljković

doi:10.5281/zenodo.18482368

February 4, 2026Open Access

Supplementary Materials for "Interpretable machine learning for cardiogram-based biometrics"

View Full Paper

Ask AI

Bookmark

Share

Key Points

To describe the preprocessing, delineation, and feature extraction pipeline for interpretable biometric identification using dual electrocardiogram and impedance cardiography signals.
Simultaneous Lead II ECG and ICG signals were recorded at 2000 Hz from N=202 healthy participants (mean age = 20.18 ± 1.88 years; 37 males, 163 females, 2 undisclosed) during relaxation baseline and anger-induction conditions.
Signals were processed using bidirectional 4th order Butterworth bandpass filters (1–40 Hz for ECG, 0.5–40 Hz for ICG), delineated for fiducial landmarks (Q, R, S, T points; B, C, X points), and ensemble-averaged over 10 consecutive beats to extract 9 averages per segment.
Extracted temporal, amplitude, morphological, and slope features were heart-rate normalized via Bazett's correction and evaluated for pairwise collinearity using Pearson and Spearman correlation coefficients.
Fiducial delineation combined with modified Pan-Tompkins detection accurately identified electrical and mechanical landmarks, including aortic valve opening (B point) and closure (X point).
Network graph analysis of feature pairs with correlation coefficients exceeding |r| > 0.7 delineated distinct structural clusters across temporal, amplitude, and angular domains.

Study Design

Type

Cross-Sectional (n=202)

Structured PICO

P

Population

202 healthy psychology students (mean age 20.2 years, 81% female) who underwent simultaneous ECG and ICG recordings during baseline relaxation and anger-induction segments.

E

Exposure

Random Forest machine learning algorithm using ECG and ICG features for biometric identification

O

Outcome

Biometric identification performance (accuracy, precision, recall, and F1 score)

This methodological study details the extraction of ECG and ICG features for machine learning-based biometric identification, highlighting that emotional states significantly alter cardiogram features and may challenge model generalization.

Limitations

Hyperparameter tuning was intentionally omitted due to computational expense.
Multicollinearity among extracted features may introduce correlation bias in feature importance estimation.
The biometric identification system assumes a closed-world scenario where unregistered individuals cannot access the system.

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

Emotion variability may limit real-world ECG/ICG biometric reliability; leaves open emotion-robust feature development for cardiogram ML.

Abstract

The repository contains Supplementary materials for the "Interpretable machine learning for cardiogram-based biometrics" paper. If you find the provided materials useful for your own research and teaching class, please cite the following references: Tanasković, I., Lazarević, L. B., Knežević, G., Milosavljević, N., Dubljević, O., Bjegojević, B., & Miljković, N. (2025). Interpretable machine learning for cardiogram-based biometrics. arXiv preprint. https://doi.org/10.48550/arXiv.2510.19775

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Tanasković et al. (2026) conducted a cross-sectional in Healthy (n=202). Anger induction vs. Baseline (relaxation) was evaluated on Differences in cardiogram-based features between baseline and anger states.

synapsesocial.com/papers/698585cb8f7c464f230097ac https://doi.org/10.5281/zenodo.18482368

View Full Paper

Ask AI

Bookmark

Share

Supplementary Materials for "Interpretable machine learning for cardiogram-based biometrics"

Key Points

Study Design

Structured PICO

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Supplementary Materials for "Interpretable machine learning for cardiogram-based biometrics"

Key Points

Study Design

Structured PICO

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading