Volume 34, Issue 7 (10-2026)
JSSU 2026, 34(7): 10451-10463 |
Back to browse issues page
Ethics code: IR.SSU.REC.1401.016
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Barzegari A, Abdulmaleki M, Mirzaei M. Prediction of Ischemic Heart Disease with A Hybrid Algorithm of Support Vector Machine (SVM) and Xgboost Using Data from Yazd Health Study (YaHS). JSSU 2026; 34 (7) :10451-10463
URL: http://jssu.ssu.ac.ir/article-1-6562-en.html
URL: http://jssu.ssu.ac.ir/article-1-6562-en.html
Full-Text [PDF 795 kb]
(35 Downloads)
| Abstract (HTML) (52 Views)
Full-Text: (13 Views)
Extended Abstract
Introduction
Ischemic heart disease (IHD) remains one of the most prevalent cardiovascular conditions and a leading cause of mortality worldwide. Early and accurate diagnosis is critical because timely intervention can significantly reduce severe complications, mortality rates, and improve patients' quality of life. Traditional diagnostic methods often require invasive procedures and impose considerable time and financial burdens on healthcare systems. In recent years, remarkable advances in machine learning and the availability of largescale health data have created new opportunities for developing intelligent prediction models. These models can analyze complex patterns within data to identify atrisk individuals and assist physicians in making more informed clinical decisions. The aim of this research was to provide an efficient instrument for the early diagnosis of ischemic heart disease and to assist clinical decisionmaking processes through the application of hybrid artificial intelligence algorithms. This study specifically sought to combine the strengths of Support Vector Machine (SVM) and XGBoost models using a Voting Classifier approach to enhance both the accuracy and efficiency of heart disease prediction.
Methods
The data used in this study were derived from the Yazd Health Study (YaHS) enrollment phase dataset, comprising 9,969 participants and 30 selective features from a total of 300, collected between 2014 and 2015. The YaHS is a populationbased cohort study of adults aged 2070 years in Yazd, Iran. Data were analyzed using SVM and XGBoost machine learning algorithms implemented in Python version 3.11.8. The dataset was divided into three parts: 70% for training, 15% for validation, and 15% for testing. The training set was used for learning model parameters, the validation set for hyperparameter tuning and model selection, and the test set for final performance evaluation on unseen data. Data splitting was performed randomly while preserving class ratios to prevent bias and data leakage.
Preprocessing steps included handling missing values through imputation methods such as mean, median, or mode replacement, and standardization using Zscore normalization to ensure that all features contributed equally to the model. Feature selection was performed using filter methods, wrapper methods, and embedded methods to identify the most relevant predictors and reduce model complexity.
For hyperparameter tuning, both SVM and XGBoost models were optimized using Grid Search with KFold CrossValidation on the training and validation sets. The SVM parameters including C and gamma, and XGBoost parameters including learning rate, n_estimators, colsample_bytree, subsample, and max_depth were systematically evaluated. The bestperforming hyperparameters were selected based on F1Score and ROC AUC metrics.
The hybrid Voting Classifier model was constructed by combining the optimized SVM and XGBoost models as base models. Soft Voting was employed, in which each model predicts the probability of sample assignment to each class, and the average of these probabilities is calculated for each class. The class with the highest average probability is selected as the final prediction. This approach typically performs better than Hard Voting because it considers more information (probabilities).
Evaluation metrics included Accuracy, Precision, Recall, F1Score, and ROC AUC. Statistical analysis was conducted using Python version 3.11.8, along with Scikitlearn version 1.4.1, NumPy version 1.26.4, Pandas version 2.2.1, and XGBoost version 2.0.3.
Results
The results demonstrated that combining the strengths of SVM and XGBoost models through a Voting Classifier approach enhanced both the accuracy and efficiency of the heart disease prediction model. The evaluation results for each model were as follows: The XGBoost model achieved an accuracy of 0.88, precision of 0.87, recall of 0.86, F1Score of 0.865, and ROC AUC of 0.91. The SVM model achieved an accuracy of 0.85, precision of 0.83, recall of 0.84, F1Score of 0.835, and ROC AUC of 0.89. The hybrid Voting Classifier model outperformed both individual models, achieving an accuracy of 0.90, precision of 0.89, recall of 0.88, F1Score of 0.885, and ROC AUC of 0.92.
Feature importance analysis using XGBoost revealed that the most important features for predicting ischemic heart disease, in order of importance, were: diabetes, family history of disease, high cholesterol, high blood pressure, age, diastolic blood pressure (DBP), systolic blood pressure (SBP), and body mass index (BMI). Based on classification, eight superior features were selected for the modeling stage. Ultimately, age was identified as one of the most important and influential factors in predicting ischemic heart disease across all models.
The comparison of model performance metrics across SVM, XGBoost, and Voting Classifier models demonstrated that the Voting Classifier consistently achieved the highest values across all metrics. ROC AUC analysis showed that the Voting Classifier achieved the highest area under the curve (0.92), followed by XGBoost (0.91) and SVM (0.89). The ROC curve analysis confirmed that the Voting Classifier model had the highest diagnostic power and was most capable of distinguishing between patients and healthy individuals, indicating that this model made the fewest diagnostic errors.
The findings of this study are consistent with previous research demonstrating the effectiveness of machine learning algorithms in predicting heart disease. Studies by Mahmoodi (2017) using SVM achieved 85% accuracy, Setiyadi et al. (2025) using SVM achieved 85% accuracy, and Gandha and Richhariya (2024) using XGBoost achieved 98.04% accuracy. The present study's hybrid approach demonstrated competitive performance while leveraging a larger, populationbased dataset from the Yazd Health Study.
Compared to the first part of this study's results, which used a combination of Random Forest, Naïve Bayes, and 20 parameters with RapidMiner software and achieved 74.51% accuracy for heart disease prediction, the present study demonstrated significantly better and higher accuracy. This improvement can be attributed to the advanced hybrid algorithm, precise preprocessing, and comprehensive hyperparameter tuning.
Conclusion
These findings suggest that implementing machine learning techniques and hybrid algorithms for early diagnosis and prediction could facilitate earlier intervention in coronary heart disease, ultimately helping to lower mortality rates. The Voting Classifier model, which combines SVM and XGBoost through Soft Voting, demonstrated superior performance compared to individual models across all evaluation metrics, including Accuracy, Precision, Recall, F1Score, and ROC AUC. Achieving high evaluation metric values (ROC AUC = 0.92) reveals the considerable potential of this model for use in early diagnosis of ischemic heart disease. Early and accurate diagnosis of ischemic heart disease is of high importance, and the developed model can serve as an efficient tool for physicians and health professionals to identify atrisk individuals, enabling timely intervention, reducing mortality, and improving patients' quality of life. Despite limitations such as interpretability and the need for validation on more diverse data, this research represents an important step toward using innovative machine learning technologies to improve community health, particularly in the field of prediction and prevention of ischemic heart diseases. Future research can further develop this field by examining more advanced ensemble algorithms, deeper interpretability methods, and using broader datasets.
Ischemic heart disease (IHD) remains one of the most prevalent cardiovascular conditions and a leading cause of mortality worldwide. Early and accurate diagnosis is critical because timely intervention can significantly reduce severe complications, mortality rates, and improve patients' quality of life. Traditional diagnostic methods often require invasive procedures and impose considerable time and financial burdens on healthcare systems. In recent years, remarkable advances in machine learning and the availability of largescale health data have created new opportunities for developing intelligent prediction models. These models can analyze complex patterns within data to identify atrisk individuals and assist physicians in making more informed clinical decisions. The aim of this research was to provide an efficient instrument for the early diagnosis of ischemic heart disease and to assist clinical decisionmaking processes through the application of hybrid artificial intelligence algorithms. This study specifically sought to combine the strengths of Support Vector Machine (SVM) and XGBoost models using a Voting Classifier approach to enhance both the accuracy and efficiency of heart disease prediction.
Methods
The data used in this study were derived from the Yazd Health Study (YaHS) enrollment phase dataset, comprising 9,969 participants and 30 selective features from a total of 300, collected between 2014 and 2015. The YaHS is a populationbased cohort study of adults aged 2070 years in Yazd, Iran. Data were analyzed using SVM and XGBoost machine learning algorithms implemented in Python version 3.11.8. The dataset was divided into three parts: 70% for training, 15% for validation, and 15% for testing. The training set was used for learning model parameters, the validation set for hyperparameter tuning and model selection, and the test set for final performance evaluation on unseen data. Data splitting was performed randomly while preserving class ratios to prevent bias and data leakage.
Preprocessing steps included handling missing values through imputation methods such as mean, median, or mode replacement, and standardization using Zscore normalization to ensure that all features contributed equally to the model. Feature selection was performed using filter methods, wrapper methods, and embedded methods to identify the most relevant predictors and reduce model complexity.
For hyperparameter tuning, both SVM and XGBoost models were optimized using Grid Search with KFold CrossValidation on the training and validation sets. The SVM parameters including C and gamma, and XGBoost parameters including learning rate, n_estimators, colsample_bytree, subsample, and max_depth were systematically evaluated. The bestperforming hyperparameters were selected based on F1Score and ROC AUC metrics.
The hybrid Voting Classifier model was constructed by combining the optimized SVM and XGBoost models as base models. Soft Voting was employed, in which each model predicts the probability of sample assignment to each class, and the average of these probabilities is calculated for each class. The class with the highest average probability is selected as the final prediction. This approach typically performs better than Hard Voting because it considers more information (probabilities).
Evaluation metrics included Accuracy, Precision, Recall, F1Score, and ROC AUC. Statistical analysis was conducted using Python version 3.11.8, along with Scikitlearn version 1.4.1, NumPy version 1.26.4, Pandas version 2.2.1, and XGBoost version 2.0.3.
Results
The results demonstrated that combining the strengths of SVM and XGBoost models through a Voting Classifier approach enhanced both the accuracy and efficiency of the heart disease prediction model. The evaluation results for each model were as follows: The XGBoost model achieved an accuracy of 0.88, precision of 0.87, recall of 0.86, F1Score of 0.865, and ROC AUC of 0.91. The SVM model achieved an accuracy of 0.85, precision of 0.83, recall of 0.84, F1Score of 0.835, and ROC AUC of 0.89. The hybrid Voting Classifier model outperformed both individual models, achieving an accuracy of 0.90, precision of 0.89, recall of 0.88, F1Score of 0.885, and ROC AUC of 0.92.
Feature importance analysis using XGBoost revealed that the most important features for predicting ischemic heart disease, in order of importance, were: diabetes, family history of disease, high cholesterol, high blood pressure, age, diastolic blood pressure (DBP), systolic blood pressure (SBP), and body mass index (BMI). Based on classification, eight superior features were selected for the modeling stage. Ultimately, age was identified as one of the most important and influential factors in predicting ischemic heart disease across all models.
The comparison of model performance metrics across SVM, XGBoost, and Voting Classifier models demonstrated that the Voting Classifier consistently achieved the highest values across all metrics. ROC AUC analysis showed that the Voting Classifier achieved the highest area under the curve (0.92), followed by XGBoost (0.91) and SVM (0.89). The ROC curve analysis confirmed that the Voting Classifier model had the highest diagnostic power and was most capable of distinguishing between patients and healthy individuals, indicating that this model made the fewest diagnostic errors.
The findings of this study are consistent with previous research demonstrating the effectiveness of machine learning algorithms in predicting heart disease. Studies by Mahmoodi (2017) using SVM achieved 85% accuracy, Setiyadi et al. (2025) using SVM achieved 85% accuracy, and Gandha and Richhariya (2024) using XGBoost achieved 98.04% accuracy. The present study's hybrid approach demonstrated competitive performance while leveraging a larger, populationbased dataset from the Yazd Health Study.
Compared to the first part of this study's results, which used a combination of Random Forest, Naïve Bayes, and 20 parameters with RapidMiner software and achieved 74.51% accuracy for heart disease prediction, the present study demonstrated significantly better and higher accuracy. This improvement can be attributed to the advanced hybrid algorithm, precise preprocessing, and comprehensive hyperparameter tuning.
Conclusion
These findings suggest that implementing machine learning techniques and hybrid algorithms for early diagnosis and prediction could facilitate earlier intervention in coronary heart disease, ultimately helping to lower mortality rates. The Voting Classifier model, which combines SVM and XGBoost through Soft Voting, demonstrated superior performance compared to individual models across all evaluation metrics, including Accuracy, Precision, Recall, F1Score, and ROC AUC. Achieving high evaluation metric values (ROC AUC = 0.92) reveals the considerable potential of this model for use in early diagnosis of ischemic heart disease. Early and accurate diagnosis of ischemic heart disease is of high importance, and the developed model can serve as an efficient tool for physicians and health professionals to identify atrisk individuals, enabling timely intervention, reducing mortality, and improving patients' quality of life. Despite limitations such as interpretability and the need for validation on more diverse data, this research represents an important step toward using innovative machine learning technologies to improve community health, particularly in the field of prediction and prevention of ischemic heart diseases. Future research can further develop this field by examining more advanced ensemble algorithms, deeper interpretability methods, and using broader datasets.
Type of Study: Original article |
Subject:
other
Received: 2025/11/8 | Accepted: 2026/05/6 | Published: 2026/10/7
Received: 2025/11/8 | Accepted: 2026/05/6 | Published: 2026/10/7
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
