Ecg Heart Disease Classification Eval

---
name: ecg-heart-disease-classification-eval
description: Evaluates the ability of deep learning models to classify heart diseases from electrocardiogram (ECG) signals. The benchmark probes multi-level feature extraction by processing ECG data hierarchically (waves, heartbeats, segments) and measures classification performance alongside model complexity and interpretability. Use when the user wants to benchmark on MIT-BIH, PTB-XL, or asks about evaluating this task. Reports accuracy.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2504.03703
  bibtex_key: [REDACTED]
  confidence: medium
---

# ecg-heart-disease-classification-eval

> Hierarchical Attention Network for Interpretable ECG-based Heart Disease Classification — Padilla Rodriguez et al. (2025) (arXiv:2504.03703, 2025)

## What this evaluates

Evaluates the ability of deep learning models to classify heart diseases from electrocardiogram (ECG) signals. The benchmark probes multi-level feature extraction by processing ECG data hierarchically (waves, heartbeats, segments) and measures classification performance alongside model complexity and interpretability.

## Datasets

- **MIT-BIH** — total ?; splits: (unstated)
- **PTB-XL** — total ?; splits: (unstated)

## Metrics

- `accuracy` **(primary)** — range: [0, 1]
  - Standard classification accuracy: the number of correctly predicted heart disease labels divided by the total number of test instances.

## Input / output format

**Input**: ECG time-series signals organized hierarchically into waves, heartbeats, and segments for model processing.

**Output**: Predicted heart disease class label.

## Scoring recipe

```python
def compute_accuracy(predictions, gold_labels):
    correct = sum(1 for p, g in zip(predictions, gold_labels) if p == g)
    return correct / len(gold_labels)
```

## Common pitfalls

- Hyperparameter search spaces are broad and not fully reported for all baseline models, making exact reproduction difficult.
- Interpretability is assessed qualitatively via attention maps rather than a standardized quantitative metric.
- Model complexity reduction claims (e.g., 19.3-fold) are mentioned but not detailed in the provided experimental text.

## Evidence (verbatim from paper)

> It demonstrates that the adapted HAN achieves near-parity with a state-of-the-art CAT-Net model in accuracy (98.55% vs. 99.14% on MIT-BIH) while reducing model complexity by up to 19.3-fold and enabling clearer interpretability through visualized attention maps highlighting clinically relevant ECG regions.

## Citation

```bibtex
@misc{padillarodriguez2025hierarchical,
  title={Hierarchical Attention Network for Interpretable ECG-based Heart Disease Classification},
  author={Padilla Rodriguez et al. (2025)},
  year={2025},
  note={arXiv:2504.03703}
}
```

- arXiv: 2504.03703