cardioinsight

##Inspiration

Cardiovascular disease is the leading cause of death worldwide, responsible for nearly 19-20 million deaths each year. Early detection of risk factors can significantly improve prevention and treatment outcomes.

The idea behind CardioInsight was to use machine learning to transform common health indicators , such as blood pressure, cholesterol, and lifestyle habits , into interpretable cardiovascular risk insights. The goal was not only to predict risk but also to help clinicians and patients understand why a certain risk level is predicted.

How I Built the Project

The system was developed using two complementary cardiovascular datasets containing clinical and lifestyle information. After cleaning the data and handling missing values, additional features such as Body Mass Index (BMI) and pulse pressure were engineered to improve model performance.

Multiple machine learning models were tested, including Logistic Regression, Random Forest, and XGBoost. XGBoost was selected as the final model due to its strong predictive accuracy.

To improve reliability, probability calibration using isotonic regression was applied. The system also integrates SHAP explainability, allowing users to see how each health factor contributes to the final risk prediction.

These insights power the Risk Simulator, which allows users to explore how lifestyle changes may influence predicted cardiovascular risk.

##Challenges

One of the main challenges was ensuring that the model remained both accurate and interpretable, which is critical for healthcare applications. Another challenge was handling noisy or unrealistic medical measurements during preprocessing.

Additionally, during evaluation, some lifestyle factors appeared to have smaller effects than expected. This occurred because the model prioritized dominant predictors such as age and systolic blood pressure, reflecting patterns learned directly from the training data.

What I Learned

This project helped deepen my understanding of machine learning pipelines, medical data preprocessing, model calibration, and explainable AI techniques. Most importantly, it highlighted the importance of transparency and interpretability when applying AI to healthcare problems.

Built With

• colab
• matplotlib
• numpy
• pandas
• python
• scikit-learn
• seaborn
• shap
• streamlit
• xgboost