Evaluating classification models in machine learning for cardiovascular risk prediction

Cardiovascular diseases, especially heart attacks, continue to pose a serious threat to public health—not just for older adults, but increasingly for children and young adults as well. Often caused by blood thickening and the formation of clots in the arteries, heart attacks can be better prevented when early warning signs and risk factors are identified in time. This study explores how machine learning can be used to predict the risk of heart stroke by analyzing patterns in patient data. We worked with a dataset of over 4,000 records, including variables like age, gender, hypertension, heart disease, marital status, occupation, where someone lives, body mass index (BMI), and smoking habits. Using this information, we trained and tested several classification models—such as Decision Trees, Binary Logistic Regression, Gaussian Naïve Bayes, Support Vector Machines (SVM), neural networks, and Logistic Regression with kernel enhancements. Each model was chosen based on its strength: Decision Trees handled categorical data well, Logistic Regression was effective for binary classification, and neural networks captured complex interactions. Among these, Binary Logistic Regression delivered the best performance, achieving an accuracy of 95.1%. Ensemble methods and Kernel Naïve Bayes followed closely with 94.9% and 94% accuracy, respectively, while Neural Networks reached 92.3%. These results highlight how machine learning can help in predicting cardiovascular risks early—potentially guiding lifestyle changes and medical interventions that save lives.