Related Experiment Video
Updated: Jun 13, 2026
07:31
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
Leakage-Safe Precision-Aware Dual-Branch FT-Transformer for Population-Scale Heart Disease Risk Prediction
Jahidul Islam1, Dristi Datta1,2, Fowzia Akhter1
1Department of Information Technology and Engineering, Sydney Metropolitan Institute of Technology (SydneyMet), Sydney, NSW 2000, Australia.
Sensors (Basel, Switzerland)
|June 12, 2026
Summary
A new Dual-Branch FT-Transformer model improves cardiovascular risk prediction for large populations. This approach enhances accuracy in screening and monitoring by addressing data imbalance and preventing information leakage.
Area of Science:
- Cardiovascular disease research
- Machine learning in healthcare
- Population health screening
Background:
- Heart disease is a leading global cause of mortality.
- Existing predictive models face challenges with imbalanced data, data leakage, and unstable precision-recall.
- Reliable population-scale risk prediction is crucial for effective screening and preventive monitoring.
Purpose of the Study:
- To propose a novel precision-aware Dual-Branch FT-Transformer framework for cardiovascular risk prediction.
- To address limitations of current machine learning models in imbalanced datasets and data leakage.
- To enhance the reliability of population-scale health monitoring.
Main Methods:
- Utilized the BRFSS-2024 dataset for cardiovascular risk prediction.
- Developed a Dual-Branch FT-Transformer architecture with specialized recall- and precision-oriented prediction heads.
- Implemented a lightweight gating mechanism trained strictly within training folds to prevent data leakage.
- Employed a strict leakage-safe 5-fold cross-validation protocol.
Main Results:
- Achieved an F1-score of 0.43, recall of 0.59, and AUPRC of 0.38 at a threshold of 0.50.
- Reduced false negatives by over 50% compared to LightGBM without significantly increasing false positives.
- Demonstrated more balanced and clinically meaningful precision-recall behavior than baseline models.
- Showcased robustness across heterogeneous population-health settings via evaluation on an independent NHANES cohort.
Conclusions:
- The proposed dual-objective learning framework offers a practical and robust solution for imbalanced tabular prediction in cardiovascular risk assessment.
- The precision-aware FT-Transformer framework enhances the reliability of population-scale screening and preventive monitoring.
- This approach effectively mitigates data leakage and improves precision-recall trade-offs in high-stakes health predictions.
