Predicting Cardiovascular Disease from Routine Clinical Measurements: A Comparative Study of Logistic Regression, Support Vector Machines, and Neural Networks

Cardiovascular disease (CVD) is one of the main causes of death worldwide, and using machine learning with basic clinical data could help identify people at risk earlier. In this project, I compared three machine learning models for predicting the presence of cardiovascular disease: Logistic Regression, Support Vector Machine (SVM), and Artificial Neural Network (MLP). I used the public Cardiovascular Disease dataset from Kaggle, which contains 70,000 patient records and 11 clinical and lifestyle features. I implemented the models using scikit-learn pipelines, with feature standardisation and hyperparameter tuning through GridSearchCV using 5-fold cross-validation. I evaluated their performance on stratified test sets using accuracy, precision, recall, F1-score, and ROC-AUC. The results were very close, with ROC-AUC scores of 0.778 for Logistic Regression, 0.786 for SVM, and 0.785 for MLP. Test accuracy ranged from 71.4% to 72.7%, showing that none of the models had a clear advantage over the others. I also used permutation importance to identify the most influential features. Systolic blood pressure was the most important predictor, followed by cholesterol, age, and weight. These findings are consistent with established clinical knowledge about cardiovascular risk factors. Finally, I discussed the main limitations of the project, including the lack of outlier treatment, differences in sample sizes and scoring metrics between models, the use of non-nested cross-validation, and the absence of external validation. This project is a research prototype for comparing machine learning approaches and is not intended for clinical diagnosis or real-world medical use.