Heart Disease Prediction Engine
ML Model for Cardiovascular Risk Assessment
Problem
Cardiovascular disease kills millions yearly yet early diagnosis in resource-limited settings relies on expensive specialist knowledge unavailable to most patients.
Solution
ML classification model trained on Cleveland Heart Disease dataset using Random Forest with feature engineering, optimized for recall to minimize missed diagnoses.
Result
87.88% accuracy. Published at IEEE INCCST'20. Cited by 13 peer-reviewed papers. Established reproducible baseline for ML cardiac risk assessment.
Impact
- ✦87.88% classification accuracy on clinical dataset
- ✦Published at IEEE Smart Gen Computing Conference
- ✦Cited by 13 peer-reviewed research papers
- ✦Established baseline for ML-based cardiac diagnostics
01The Problem
Cardiovascular disease is the leading cause of death globally, yet early diagnosis remains inconsistent | especially in resource-limited settings where specialist access is rare. The goal was to build a machine learning model that could predict cardiovascular risk from standard clinical measurements available in any basic health checkup, with high enough accuracy to be clinically useful.
02The Dataset & Approach
We used the Cleveland Heart Disease dataset from the UCI repository | 303 patient records with 14 clinical features including age, cholesterol, blood pressure, chest pain type, and ECG results. The target variable was binary: presence or absence of heart disease. I ran exploratory data analysis to understand feature distributions, correlations, and missing value patterns before any modelling.
03Technical Approach
I tested multiple classification algorithms | Logistic Regression, Decision Trees, Random Forest, K-Nearest Neighbours, and SVM | using cross-validation to avoid overfitting. Feature engineering was the most impactful step: creating interaction features between correlated variables and applying feature selection to remove noise. Random Forest with tuned hyperparameters achieved the highest accuracy. The final model was evaluated on precision, recall, and F1-score | not just accuracy | since false negatives (missing a sick patient) are far more costly than false positives.
04Key Decisions
The decision to optimise for recall over precision was deliberate | in a diagnostic context, missing a sick patient is worse than a false alarm. The final model was tuned to minimise false negatives even at the cost of some precision. I also documented the full model card including training data characteristics, performance metrics, and known limitations | essential for any model intended for clinical decision support.
05Outcome
The model achieved 87.88% accuracy with strong recall, was published at the IEEE International Conference on Smart Generation Computing, and has since been cited by 13 peer-reviewed papers. The work established a reproducible baseline for ML-based cardiac risk assessment that subsequent researchers have built on.