Predicting Cardiovascular Disease Using Machine Learning

CardioDisease_Prediction is a machine learning project aimed at predicting cardiovascular disease (CVD) risk using the publicly available cardio_train dataset. This project was completed as part of the Foundations of Statistical Analysis & Machine Learning CSC1181 course project by Team 52.

---

Project Overview

Cardiovascular disease is one of the leading causes of death worldwide. Early prediction of risk can guide timely interventions. This project focuses on the cardio_train dataset containing 70,000 anonymized patient records and builds machine learning models to predict cardiovascular disease status, where:

• 0 indicates healthy
• 1 indicates the presence of a cardiovascular disease

The project workflow included:

1. Data Preprocessing

   • Loaded dataset using pandas and inspected for missing values and duplicates
   • Removed physiologically unrealistic values for blood pressure, height, and weight
   • Created new features: age_years from age in days and BMI from height and weight
   • Standardized continuous variables (height, weight, blood pressure) using StandardScaler
   • Split data into 80% training and 20% test sets using stratified sampling

2. Models Implemented

   • Logistic Regression: Baseline linear model providing feature influence interpretation
   • Random Forest Classifier: Ensemble model robust to noise and capable of capturing non-linear patterns
   • Gradient Boosting (GB): Sequential ensemble improving predictive performance iteratively
   • Extreme Gradient Boosting (XGBoost): Optimized GB with GPU acceleration and hyperparameter tuning for efficiency and accuracy

3. Validation and Evaluation

   • Applied 5-fold stratified cross-validation on XGBoost to ensure stable performance
   • Evaluated models using accuracy, F1-score, confusion matrices, and ROC-AUC
   • Final test set metrics for XGBoost with GPU:
     • Accuracy: 0.7293
     • F1 Score: 0.7140
     • Confusion Matrix:

       [[5370 1562]
        [2152 4635]]
       

   • Feature importance ranked top predictors as BMI, systolic and diastolic blood pressure, age, height, and weight

4. Insights

   • Boosting models outperformed simpler approaches, with XGBoost (using GPU) offering best performance and efficiency
   • Feature importance aligns with known clinical risk factors, confirming model learned meaningful patterns
   • The model generalizes well, with consistent cross-validation results (mean CV F1 ≈ 0.721, SD ≈ 0.006)

---

Ethical and Legal Considerations

• Dataset is anonymized and contains no personal identifiers
• Model is intended for academic purposes only and cannot replace clinical judgment
• Real-world deployment would require GDPR compliance, strong data security, and clear communication of model limitations

Team Members & Contributions

Name	Role	Key Responsibilities	Gmail ID
Achal Nanjundamurthy	Model Building, Cross-validation, Research	Implemented and trained models, performed cross-validation and hyperparameter tuning, produced final metrics and plots	achalnm02@gmail.com
Khushi Bijkal	Pre-processing, Visualization, Reporting	Handled data preprocessing, exploratory data analysis (EDA) visualizations, interpreted results, and finalized the report	khushi.n.bijkal@gmail.com
Suman Neupane	Model Building, Error Analysis, Research	Supported model building, performed error and metric analysis, gathered related research papers and references	sumanneupane99@gmail.com
Durga Prasad Narsing	Analysis, Reporting	Provided statistical and ethical/legal write-up, helped interpret results and compile the final report	durgaprasadnarsing1728@gmail.com