ML Classifiers for Cardiac Arrhythmia Detection from PPG: Algorithms, Accuracy & Clinical Evidence

Classical Machine Learning Classifiers

Support Vector Machines (SVM)

SVMs with radial basis function (RBF) kernels have been widely applied to PPG arrhythmia detection, particularly in the early literature. Typical feature sets combine IBI statistics computed over fixed windows. The key advantage of SVMs is their strong performance with relatively small training datasets, which was important given the limited availability of labeled PPG arrhythmia data in earlier studies.

Random Forests and Gradient Boosting

Ensemble tree methods have become the dominant classical ML approach for PPG arrhythmia classification. Random Forests provide built-in feature importance ranking, which aids in understanding which signal characteristics drive classification decisions.

Signal quality handling matters as much as the classifier. Tarniceriu et al. (2018) detected AF from windows of 20 consecutive wrist-PPG inter-beat intervals in 29 post-operative patients, with 98.45% sensitivity and 99.13% specificity, but a decision could only be made for about 76% of the time because unreliable intervals were discarded (doi:10.1109/EMBC.2018.8513197).

Gradient boosting classifiers such as XGBoost are also used for these feature sets, including for multi-class problems that separate normal sinus rhythm, AF, and other arrhythmias.

Hidden Markov Models (HMMs)

HMMs are particularly well-suited for arrhythmia detection because they model temporal sequences of states, capturing the transitions between normal rhythm and arrhythmia. Bonomi et al. (2018) used a Markov model to estimate the probability of AF from irregular patterns in wrist-PPG inter-pulse intervals, with motion-corrupted intervals discarded. AF was detected with 97% sensitivity in 20 cardioversion patients and 93% sensitivity in 40 Holter patients, both with 100% specificity, although no classification was available for 44% of the Holter monitoring period because of motion artifacts (doi:10.1161/JAHA.118.009351).

Deep Learning Architectures

1D Convolutional Neural Networks

1D CNNs operating on raw or minimally processed PPG waveforms have become the most popular deep learning approach for arrhythmia detection. Their key advantage is automatic feature learning, eliminating the need for hand-crafted feature engineering.

Architecture patterns. Successful 1D CNN architectures for PPG arrhythmia classification typically use 4-8 convolutional layers with increasing filter counts (32, 64, 128, 256), batch normalization, ReLU activation, and max pooling. The input is typically a 10-30 second PPG segment sampled at 64-256 Hz. Global average pooling before the output layer reduces the parameter count and provides some regularization.

Shen et al. (2019) trained a 50-layer CNN on more than 4,000 hours of annotated wrist PPG recorded in free-living conditions and achieved a test AUC of 0.95 for AF despite motion artifacts (doi:10.1145/3292500.3330657). Tison et al. (2018) trained a deep neural network on smartwatch heart rate and step count data from 9,750 participants; it reached a C statistic of 0.97 (sensitivity 98.0%, specificity 90.2%) against 12-lead ECG in 51 cardioversion patients, but only 0.72 in an exploratory ambulatory analysis (doi:10.1001/jamacardio.2018.0136).

Recurrent Neural Networks and LSTMs

Recurrent architectures are a natural fit for sequential PPG data, as they can model temporal dependencies across extended time windows. Long Short-Term Memory (LSTM) networks have been applied to both IBI sequences and raw PPG waveforms.

Shashikumar et al. (2017) applied deep learning to wearable PPG and accelerometry for AF monitoring (doi:10.1109/BHI.2017.7897225). Bidirectional architectures let a model use both past and future context when classifying each segment.

For wearable deployment, LSTM models face challenges due to their sequential computation requirement (each time step depends on the previous hidden state), which limits parallelization and increases latency. Gated Recurrent Units (GRUs) offer a simpler alternative with comparable performance and reduced computational cost.

Hybrid CNN-LSTM Models

Combining CNNs for local feature extraction with LSTMs for temporal modeling has proven particularly effective. The CNN layers extract morphological features from individual beats or short segments, while the LSTM layers model the temporal evolution of these features to detect sustained arrhythmias.

Torres-Soto and Ashley (2020) developed DeepBeat, a multitask model that jointly assesses signal quality and detects AF from wearable PPG. Pretrained on about one million simulated signals with convolutional denoising autoencoders and fine-tuned on over 500,000 labeled signals from more than 100 individuals, it improved AF detection F1 from 0.54 (single-task) to 0.96, and reached sensitivity 0.98 and specificity 0.99 in a prospective ambulatory replication cohort (doi:10.1038/s41746-020-00320-4).

Transformer Models

Attention-based transformer architectures have recently been applied to PPG arrhythmia detection, leveraging their ability to capture long-range dependencies without the sequential processing bottleneck of RNNs. Their computational cost is higher than CNN-based approaches, which makes real-time wearable deployment challenging.

Clinical Validation Studies

Apple Heart Study

The Apple Heart Study (Perez et al., 2019) enrolled 419,297 participants wearing Apple Watch devices with a PPG-based irregular rhythm notification algorithm. Over 8 months of monitoring, 0.52% of participants received an irregular pulse notification. Among the 450 notified participants who returned analyzable ECG patches, 34% had AF. The positive predictive value was 0.84 for observing AF on the ECG simultaneously with a subsequent irregular pulse notification (doi:10.1056/NEJMoa1901183).

This study established the feasibility of population-scale AF screening using wearable PPG but also highlighted the challenge of low positive predictive value when screening a low-prevalence population. The 34% AF confirmation rate among notified participants means that 66% received false-positive notifications, underscoring the need for high specificity in screening algorithms.

Huawei Heart Study

The Huawei Heart Study (Guo et al., 2019) enrolled 187,912 participants using Huawei smartwatches and bands. The PPG-based algorithm flagged 424 participants (0.23%) with suspected AF; of the 262 who were effectively followed up, 227 (87.0%) were confirmed to have AF, and the positive predictive value of the PPG signals was 91.6% (doi:10.1016/j.jacc.2019.08.019).

Fitbit Heart Study

The Fitbit Heart Study (Lubitz et al., 2022) enrolled 455,699 participants and used a PPG algorithm that analyzed overlapping 5-minute pulse windows (tachograms) and required 11 consecutive irregular tachograms before a notification. Among 1,057 notified participants with analyzable ECG patches, 32.2% had AF; when another detection occurred during patch monitoring, the positive predictive value for concurrent AF was 98.2% (doi:10.1161/CIRCULATIONAHA.122.060291).

Deployment Considerations

Computational Constraints

Wearable devices impose strict computational limits: ARM Cortex-M class processors, limited RAM (32-256 KB), and battery life constraints. Models deployed on these platforms must be compressed through quantization (INT8 or binary), pruning, and knowledge distillation. Successful deployments have achieved model sizes under 100 KB while maintaining AUC > 0.94 for AF detection. For more on embedded PPG signal processing, see our algorithms documentation.

Signal Quality Gating

Real-world PPG data contains substantial periods of motion artifact, sensor disconnection, and low perfusion. All production arrhythmia detection systems include signal quality assessment (SQA) as a gating mechanism. Only segments that pass SQA thresholds are forwarded to the arrhythmia classifier. This approach trades reduced sensitivity (arrhythmia episodes during motion may be missed) for improved specificity (reduced false positives from motion artifact).

Typical SQA features include: peak-to-peak amplitude thresholds, perfusion index minimums, template matching scores, and accelerometer-based motion intensity thresholds. Well-designed SQA can reject 30-60% of ambulatory PPG data while retaining >95% of arrhythmia-containing segments.

Multi-Class vs. Binary Classification

Most deployed systems use binary classification (AF vs. non-AF) because AF is the most clinically impactful arrhythmia detectable by PPG and has the strongest evidence base. Multi-class classification (AF, PVC, PAC, flutter, tachycardia, bradycardia) remains primarily a research endeavor, with reported accuracies 10-15% lower than binary AF detection due to the increased complexity and the hemodynamic similarity between certain arrhythmia types.

Conclusion

Machine learning classification of cardiac arrhythmias from PPG has matured from a laboratory demonstration to a clinically deployed capability for atrial fibrillation screening, validated in studies enrolling over one million participants collectively. The combination of robust feature engineering, deep learning architectures, and rigorous signal quality gating has produced systems with sensitivity exceeding 95% and specificity above 97% for AF detection under controlled conditions.

The next frontiers include multi-class arrhythmia discrimination, detection of lower-prevalence rhythm disorders, improved performance during physical activity, and integration with on-demand ECG confirmation. For researchers entering this field, the combination of publicly available PPG databases (MIMIC, IEEE SP Cup), open-source beat detection algorithms, and established feature sets provides a solid foundation. For current clinical applications and regulatory status, see our guide to wearable AF detection and our overview of health conditions monitorable by PPG.

References

• Elgendi, M., Norton, I., Brearley, M., Abbott, D., & Schuurmans, D. (2013). Systolic peak detection in acceleration photoplethysmograms measured from emergency responders in tropical conditions. PLoS ONE, 8(10), e76585. https://doi.org/10.1371/journal.pone.0076585
• Dash, S., Chon, K. H., Lu, S., & Raeder, E. A. (2009). Automatic real time detection of atrial fibrillation. Annals of Biomedical Engineering, 37(9), 1701-1709. https://doi.org/10.1007/s10439-009-9740-z
• Sarkar, S., Ritscher, D., & Mehra, R. (2008). A detector for a chronic implantable atrial tachyarrhythmia monitor. IEEE Transactions on Biomedical Engineering, 55(3), 1219-1224. https://doi.org/10.1109/TBME.2007.903707
• Tarniceriu, A., et al. (2018). The accuracy of atrial fibrillation detection from wrist photoplethysmography: a study on post-operative patients. 2018 40th Annual International Conference of the IEEE EMBC, 1-4. https://doi.org/10.1109/EMBC.2018.8513197
• Bonomi, A. G., et al. (2018). Atrial fibrillation detection using a novel cardiac ambulatory monitor based on photo-plethysmography at the wrist. Journal of the American Heart Association, 7(15), e009351. https://doi.org/10.1161/JAHA.118.009351
• Tison, G. H., et al. (2018). Passive detection of atrial fibrillation using a commercially available smartwatch. JAMA Cardiology, 3(5), 409-416. https://doi.org/10.1001/jamacardio.2018.0136
• Shen, Y., et al. (2019). Ambulatory atrial fibrillation monitoring using wearable photoplethysmography with deep learning. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 1909-1916. https://doi.org/10.1145/3292500.3330657
• Shashikumar, S. P., Shah, A. J., Li, Q., Clifford, G. D., & Nemati, S. (2017). A deep learning approach to monitoring and detecting atrial fibrillation using wearable technology. 2017 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), 141-144. https://doi.org/10.1109/BHI.2017.7897225
• Torres-Soto, J., & Ashley, E. A. (2020). Multi-task deep learning for cardiac rhythm detection in wearable devices. npj Digital Medicine, 3, 116. https://doi.org/10.1038/s41746-020-00320-4
• Perez, M. V., et al. (2019). Large-scale assessment of a smartwatch to identify atrial fibrillation. New England Journal of Medicine, 381(20), 1909-1917. https://doi.org/10.1056/NEJMoa1901183
• Guo, Y., et al. (2019). Mobile photoplethysmographic technology to detect atrial fibrillation. Journal of the American College of Cardiology, 74(19), 2365-2375. https://doi.org/10.1016/j.jacc.2019.08.019
• Lubitz, S. A., et al. (2022). Detection of atrial fibrillation in a large population using wearable devices: the Fitbit Heart Study. Circulation, 146(19), 1415-1424. https://doi.org/10.1161/CIRCULATIONAHA.122.060291