ECG Classification on PTB-XL: A Data-Centric Approach with Simplified CNN-VAE

Naqcho Ali Mehdi; Amir Ali

cs.LGMar 8, 2026

ECG Classification on PTB-XL: A Data-Centric Approach with Simplified CNN-VAE

Authors: Naqcho Ali Mehdi, Amir Ali

Organizations: NED University of Engineering and Technology, Karachi, Pakistan

Abstract

Automated electrocardiogram (ECG) classification is essential for early detection of cardiovascular diseases. While recent approaches have increasingly relied on deep neural networks with complex architectures, we demonstrate that careful data preprocessing, class balancing, and a simplified convolutional neural network combined with a variational autoencoder (CNN-VAE) architecture can achieve competitive performance with significantly reduced model complexity. Using the publicly available PTB XL dataset, we achieve 87.01% binary accuracy and 0.7454 weighted F1-score across five diagnostic classes (CD, HYP, MI, NORM, STTC) with only 197,093 trainable parameters. Our work emphasises the importance of data-centric machine learning practices over architectural complexity, demonstrating that systematic preprocessing and balanced training strategies are critical for medical signal classification. We identify challenges in minority class detection (particularly hypertrophy) and provide insights for future improvements in handling imbalanced ECG datasets. Index Terms: ECG classification, convolutional neural networks, class balancing, data preprocessing, variational autoencoders, PTB-XL dataset

Figures & tables

Metric	Value
Test Loss	0.4219
Bin. Accuracy	0.8701
Precision	0.7354
Recall	0.7640
AUC	0.8958
Hamming Loss	0.1299

Table 1: Overall Test Set Performance

Class	Prec.	Rec.	F1	Sup.
CD	0.728	0.699	0.713	498
HYP	0.576	0.502	0.537	263
MI	0.730	0.678	0.703	553
NORM	0.796	0.910	0.849	964
STTC	0.695	0.780	0.735	523
W. Avg	0.731	0.764	0.745	2801

Table 2: Per-Class Classification Metrics

Figure 1: Confusion matrix for NORM class showing excellent recall (91%) with 877 true positives out of 964 actual normal cases, indicating the model’s strong ability to correctly identify healthy ECGs.

Figure 2: Confusion matrix for CD (Conduction Disturbances) class showing moderate performance with 348 true positives and 150 false negatives, reflecting the subtle ECG changes associated with conduction abnormalities.

Figure 3: Confusion matrix for HYP (Hypertrophy) class revealing the model’s primary weakness, with 131 false negatives (50% missed cases) out of 263 actual hypertrophy cases, highlighting the challenge of detecting this subtle condition.

Figure 4: Confusion matrix for MI (Myocardial Infarction) class showing balanced performance with 375 true positives and 178 false negatives, indicating reasonable but improvable detection of acute cardiac events.

Figure 5: Confusion matrix for STTC (ST/T-wave Changes) class demonstrating good recall with 408 true positives and relatively low false negatives (115), showing effective detection of ischemic changes.

Figure 6: Aggregate confusion matrix across all classes, visualizing the overall classification performance and error distribution pattern.

Class	TN	FP	FN	TP
NORM	1014	225	87	877
HYP	1843	97	131	132
MI	1511	139	178	375
STTC	1501	179	115	408
CD	1575	130	150	348

Table 3: Confusion Matrix Summary (TN, FP, FN, TP)

Figure 7: Training curves showing loss and accuracy convergence over 69 epochs. Upper left: Training loss decreases monotonically while validation loss plateaus around epoch 40-50. Upper right: Both training and validation accuracy increase with validation lagging training. Lower left: Training precision exceeds validation, consistent with class weight effects. Lower right: Validation recall stabilizes around 76%, indicating learned balance between sensitivity and specificity.

Method	Acc.	F1
ResNet-50	82.3%	—
CNN-VAE	87.0%	0.745
Modern Apps	82–88%	0.70–0.78

Table 4: Comparison to Related Work

Explore similar work

ECG-Lens: Benchmarking ML & DL Models on PTB-XL Dataset
Apr 17, 2026Saloni Garg, Ukant Jadia, Amit Sagtani +1Electrocardiogram Classification Convolutional Neural Networks
Sir Padampat Singhania University Udaipur, Rajasthan, India · The LNM Institute of Information Technology, Jaipur, India
Boosting ECG Classification Performance by Pre-training with Synthesized Data
Jun 9, 2026Naoki Nonaka, Jun SeitaSynthetic Data Pretraining Synthetic-to-Real Domain Adaptation
Advanced Data Science Project, RIKEN Information R&D and Strategy Headquarters
ECG-NAT: A Self-supervised Neighborhood Attention Transformer for Multi-lead Electrocardiogram Classification
May 13, 2026Mahsa Gazeran, Sayvan Soleymanbaigi, Fatemeh Daneshfar +2Self-Supervised Time Series Representation Learning Transformer Attention
Department of Computer Engineering, University of Kurdistan, Sanandaj, Iran · Department of Mathematics and Operational Research, University of Mons, Mons, Belgium

Apr 17, 2026J/K move · Enter open · S save

ECG-Lens: Benchmarking ML & DL Models on PTB-XL Dataset

Saloni Garg, Ukant Jadia, Amit Sagtani, Kamal Kant Hiran

Sir Padampat Singhania University Udaipur, Rajasthan, India · The LNM Institute of Information Technology, Jaipur, India

Electrocardiogram Classification Convolutional Neural Networks

Open paper AskPDF arXiv

Automated classification of electrocardiogram (ECG) signals is a useful tool for diagnosing and monitoring cardiovascular diseases. This study compares three traditional machine learning algorithms (Decision Tree Classifier, Random Forest Classifier, and Logistic Regression) and three deep learning models (Simple Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Complex CNN (ECGLens)) for the classification of ECG signals from the PTB-XL dataset, which contains 12-lead recordings from normal patients and patients with various cardiac conditions. The DL models were trained on raw ECG signals, allowing them to automatically extract discriminative features. Data augmentation using the Stationary Wavelet Transform (SWT) was applied to enhance model performance, increase the diversity of training samples, and preserve the essential characteristics of the ECG signals. The models were evaluated using multiple metrics, including accuracy, precision, recall, F1-score, and ROC-AUC. The ECG-Lens model achieved the highest performance, with 80% classification accuracy and a 90% ROC-AUC. These findings demonstrate that deep learning architectures, particularly complex CNNs substantially outperform traditional ML methods on raw 12-lead ECG data, and provide a practical benchmark for selecting automated ECG classification models and identifying directions for condition-specific model development.

Abstract

Explore similar work

ECG-Lens: Benchmarking ML & DL Models on PTB-XL Dataset

Boosting ECG Classification Performance by Pre-training with Synthesized Data

ECG-NAT: A Self-supervised Neighborhood Attention Transformer for Multi-lead Electrocardiogram Classification