Heart Rate Modeling and Arrhythmia Detection Using Deep Learning Neural Networks: A Dual-Task TensorFlow Pipeline for Cardiac Health Prediction

Cardiovascular disease remains the leading cause of mortality worldwide. This study presents a complete dual-task deep learning pipeline addressing two complementary cardiac prediction problems. Part A develops a regression neural network for continuous heart rate prediction from ten lifestyle and physiological features including age, BMI, exercise duration, sleep hours, stress level, and caffeine intake, achieving a Mean Absolute Error of 3.80 BPM and R2 Score of 0.8682. Part B develops a multi-class classification neural network for arrhythmia detection across five cardiac rhythm categories including Normal, Atrial Fibrillation, Bradycardia, Tachycardia, and Premature Ventricular Contraction, achieving 99.5% accuracy and a macro-averaged ROC-AUC of 0.9999. Both models incorporate Batch Normalization, Dropout regularization, EarlyStopping, and ReduceLROnPlateau callbacks implemented in TensorFlow and Keras. Source code, trained model artifacts, and figures are publicly available on GitHub (https://github.com/randomthingsonlineatsk-cloud/heart-rate-neural-network), Kaggle Part A (https://www.kaggle.com/code/shagufakhan/heart-rate-regression), Kaggle Part B (https://www.kaggle.com/code/shagufakhan/arrhythmia-classification), and archived on Zenodo (DOI: 10.5281/zenodo.20388876).