Quantum-Swarm Hybrid Framework for Automated Hyperparameter Optimization in Boosting Algorithms for Cardiac Disease Prediction

Résumé

Heart disease is one of the major causes of death across the globe; therefore, correct models of diagnostic prediction are required to aid clinical decisions. This work entails a new use of the Quantum Particle Swarm Optimization (QPSO) in optimizing hyperparameters of three different boosters-XGBoost, AdaBoost and Gradient Boosting on the prediction of heart diseases using clinical and demographic data. QPSO uses quantum mechanics concepts such as superposition and tunneling to find the best hyperparameter settings better than the classical methods of optimization by avoiding local optima. QPSO was applied on a set of 11 cardiac features (age, sex, blood pressure, cholesterol, fasting blood sugar, resting ECG, maximum heart rate, exercise-induced angina, ST depression, and ST segment slope) with binary disease classification and 30 particles in 150 iterations producing 4500 model trainings per algorithm. QPSO + XGBoost performed well with the highest accuracy of 85.33%, 87.25% recall (which is above the clinical target of 85), 86.41% precision, F1-score 0.8683 and ROC-AUC 0.9201 with 53 out of 61 disease cases identified correctly. Of the established cardiac risk factors, the feature importance analysis found both the ST_Slope and maximum heart rate to have top-tier feature predictions with an average rank of 1.67. AdaBoost converged quickly after iteration 20 with the best fitness of 0.7843 focusing on physiological measures by reweighting the samples whereas XGBoost and Gradient Boosting converged at iteration 60 and 100 respectively with various hyperparameter approaches. The study shows that QPSO is applicable in the identification of optimal configurations of algorithms that can outperform default parameters, to improve the use of medical machine learning in detecting cardiovascular disease. Findings favour algorithm-based clinical diagnosis through suitable integration and optimal threshold to deploy in health care.