Predicting hypertension risk using machine learning, deep learning, and quantum neural network models: a comparative evaluation

Hypertension is a major public health problem and an important risk factor for cardiovascular disease. Early identification of patients at risk of hypertension can support timely prevention and clinical assessment. This study developed and compared statistical, machine learning, deep learning, and quantum neural network (QNN) models for hypertension prediction using clinical data from 624 patients at District Headquarters Hospital, Faisalabad, Pakistan. The proposed framework involved data preprocessing, K-nearest neighbor imputation, chi-square-based feature selection, feature scaling, and synthetic minority over-sampling technique (SMOTE) for class balancing. Logistic regression was used as a statistical reference, while SVM, random forest, XGBoost, Naive Bayes, ANN, SVM with PCA, DNN, and QNN were evaluated. Stratified 10-fold cross-validation was used to assess model performance using accuracy, F-measure, AUC, precision, recall, specificity, and maximum probability of correct decision (MPCD). The DNN model achieved the highest accuracy (95.77%), while the QNN achieved the highest AUC (95.40%), precision (96.26%), and recall (94.76%). Naive Bayes showed the lowest accuracy (83.40%). The QNN was included as an exploratory hybrid quantum-classical approach using a simulated quantum circuit, and the results do not imply a quantum advantage over classical models. Overall, the evaluated models showed promising predictive performance for hypertension classification in the study population. Further evaluation using independent populations, external validation, and prospective clinical studies is needed before considering their use in routine clinical practice.