Deep Learning-Based Intelligent Heart Rate Prediction and Arrhythmia Detection System

This project is a complete deep learning solution for cardiovascular health, combining a regression model to predict resting heart rate and a multi-class neural network to detect five common arrhythmia types with an accuracy of up to 99.5%. Built using the TensorFlow and Keras frameworks, it features a dual-module design covering key dimensions of cardiovascular health monitoring, providing a scalable technical solution for personal health monitoring and clinical auxiliary diagnosis.

Cardiovascular diseases are one of the leading causes of death globally. Early identification of heart rate abnormalities and arrhythmias is crucial for preventing serious complications. Traditional monitoring relies on professional equipment; this project provides a scalable solution through deep learning technology. Developed by independent researchers using the TensorFlow/Keras framework, it includes dual modules for resting heart rate prediction (regression) and arrhythmia detection (classification), covering both continuous numerical indicator prediction and discrete pathological state recognition.

Module A: Resting Heart Rate Regression Prediction

Input features include age, weight, height, BMI, activity level, daily steps, sleep duration, exercise duration, stress level, and caffeine intake. Test set performance: MAE=3.80 BPM, RMSE=4.81 BPM, R²=0.8682. BMI and stress are strongly correlated factors for increased heart rate.

Module B: Multi-Class Arrhythmia Detection

Based on ECG-derived features (RR interval, heart rate, PR interval, QRS duration, QT interval, heart rate variability indicators, frequency domain features), it detects five types (normal, atrial fibrillation, bradycardia, tachycardia, ventricular premature beats). Performance: accuracy 99.5%, ROC-AUC=0.9999.

1. Patients with atrial fibrillation have the highest RR interval variability (SDNN), which aligns with the pathological feature of absolute arrhythmia;
2. Patients with tachycardia have significantly lower RMSSD and pNN50, indicating impaired autonomic regulation of the heart and reduced parasympathetic activity;
3. BMI and stress levels are key factors for elevated resting heart rate, emphasizing the importance of lifestyle interventions. These findings support the model's interpretability and provide data-driven evidence for clinical practice.

• Reproducibility: Provides a complete Kaggle Notebook link, which can be run without local configuration;
• Visualization: Generates 8 sets of charts (feature distribution, correlation heatmap, model evaluation, etc.);
• Dependency management: Clear dependencies in requirements.txt;
• Academic compliance: Provides DOI citation identifier.

Application Scenarios:

• Wearable device integration: Integrate with smartwatch sensors for real-time monitoring and early warning;
• Clinical auxiliary diagnosis: Provide second opinions for technicians and experts;
• Health risk assessment: Predict risks based on lifestyle data to support personalized interventions.

Expansion Directions: Introduce time-series models (LSTM/Transformer) to process continuous ECG signals, integrate multi-modal data (blood oxygen, blood pressure), and develop an edge computing version to support low-latency inference.

This project demonstrates a complete deep learning solution for cardiovascular health, forming a full chain from data preprocessing to visualization. The 99.5% arrhythmia classification accuracy and R² regression score of 0.87 prove its effectiveness. For medical AI developers, the project provides a clear code structure and detailed documentation, serving as a learning example; meanwhile, the analysis of feature importance and clinical findings reflects the value of combining data science with medicine.