• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Monday, November 16, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Optimized Machine Learning Model Predicts Coronary Artery Disease with High Accuracy

by
November 16, 2026
in Technology
Reading Time: 6 mins read
0
Optimized Machine Learning Model Predicts Coronary Artery Disease with High Accuracy

Optimized Machine Learning Model Predicts Coronary Artery Disease with High Accuracy

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

A recent study has developed an advanced machine learning framework that significantly enhances the non-invasive prediction of coronary artery disease (CAD). By integrating ensemble-based feature selection with a support vector machine (SVM) classifier tuned via grid search, researchers achieved a diagnostic accuracy of 96.86% on the Z-Alizadeh Sani dataset, which contains 303 patient samples. This methodology addresses two primary challenges in clinical data analysis: the high dimensionality of cardiovascular risk factors and the inherent class imbalance commonly found in medical datasets, where the number of healthy controls is significantly lower than that of diagnosed CAD patients.

To overcome the limitations of standard diagnostic procedures like coronary angiography, which is costly, time-consuming, and carries procedural risks, researchers increasingly rely on computer-aided decision-making tools. This research employed a rigorous nested cross-validation framework to prevent data leakage, ensuring that preprocessing steps such as feature selection and oversampling were conducted independently within each fold. By utilizing a hybrid feature selection approach that combines the importance scores from both Random Forest and XGBoost algorithms, the study successfully reduced the dimensionality from 54 original variables to 42 highly informative features. This refined feature set allows for more focused computational analysis while preserving critical diagnostic information necessary for accurate CAD identification.

A critical component of this study was the systematic evaluation of three distinct oversampling techniques designed to mitigate class imbalance: Synthetic Minority Over-sampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), and Conditional Tabular Generative Adversarial Network (CT-GAN). The comparative analysis revealed that ADASYN, which generates synthetic samples adaptively based on the difficulty of learning specific instances near decision boundaries, significantly outperformed the other methods. ADASYN achieved superior classification outcomes, including 95.24% sensitivity and a perfect AUC of 1.00, demonstrating its effectiveness in handling complex cardiovascular data configurations compared to the generative adversarial network approach, which struggled with the limited sample size and structural complexity of the dataset.

The classification performance of five different algorithms—Random Forest, Logistic Regression, XGBoost, Multi-Layer Perceptron (MLP), and SVM—was systematically optimized using both grid search and random search methods. The results highlighted that the SVM model, when coupled with ensemble feature selection and ADASYN-based oversampling, provided the most robust clinical-grade performance. This combination effectively stabilized the training process and improved the model’s ability to generalize to unseen data. When compared to traditional baseline models, the proposed methodology demonstrated statistically significant improvements, as confirmed by McNemar’s tests and paired t-tests, particularly in accurately distinguishing between CAD patients and healthy individuals.

Beyond predictive performance, the study prioritized clinical interpretability by implementing SHAP (SHapley Additive exPlanations) analysis to demystify the machine learning output. This technique provided a transparent hierarchy of risk factors, revealing that typical chest pain was the most influential predictor with a SHAP value of 0.567. Other major contributors included atypical chest pain, left-ventricular ejection fraction (EF-TTE), advancing age, and hypertension. These findings align with established cardiovascular knowledge, providing clinicians with clear justifications for the model’s predictions and enhancing the potential for integration into real-world clinical decision support systems where transparency is essential for medical adoption.

The study further validated the robustness of the proposed methodology by evaluating the model across four publicly available cardiovascular datasets, including those from Cleveland, Statlog, and the Long Beach VA. While the accuracy varied across these diverse patient populations due to differences in sample size, feature dimensionality, and specific healthcare settings, the model maintained reasonable generalizability. These results underscore the complexity of adapting CAD diagnosis models to different clinical environments and highlight the necessity for population-specific models or transfer learning frameworks to account for regional differences in medical data distributions.

Computational efficiency was another significant finding, with the model achieving an average training time of approximately 0.91 seconds per fold. This rapid processing time is a crucial attribute for clinical applications, where both high diagnostic accuracy and speed are required for effective patient management and resource allocation during consultations. By bridging the gap between sophisticated algorithmic techniques and the practical requirements of a clinical setting, the study establishes a new benchmark for non-invasive heart disease prediction that reduces reliance on invasive diagnostic techniques.

Despite these promising results, the authors acknowledge several limitations, primarily the relatively small sample size of the Z-Alizadeh Sani dataset. Although multi-dataset validation was performed, the potential for overfitting in high-dimensional feature spaces remains a concern, particularly given the perfect performance metrics observed in certain configurations. The researchers emphasize that the high specificity and precision achieved may require further validation on much larger, multi-center cohorts to fully confirm clinical utility across broader demographic groups.

Looking toward future advancements, the research suggests that incorporating metaheuristic optimization algorithms—such as Particle Swarm Optimization or Greylag Goose Optimization—could further refine feature selection and hyperparameter tuning. Moreover, future integration with Electronic Health Record (EHR) systems and the application of federated learning could facilitate real-time, privacy-preserving collaborative training across different hospitals. These developments would allow for the creation of more adaptive and universally applicable diagnostic models that could significantly contribute to the reduction of global cardiovascular mortality rates through earlier intervention strategies.

In conclusion, the proposed Ensemble FS + ADASYN + SVM framework offers a comprehensive, interpretable, and computationally efficient solution for the automated detection of CAD. By systematically addressing class imbalance and feature selection while maintaining robust statistical validation, the approach demonstrates how well-configured machine learning models can complement clinical expertise. As researchers continue to refine these tools, the path toward integrating artificial intelligence into cardiology practice becomes increasingly clear, promising faster, more accurate, and non-invasive diagnostic capabilities for patients worldwide.

The urgent clinical need for reliable computer-aided diagnostic tools is underscored by international cardiovascular mortality trends. According to World Health Organization estimates cited in the investigation, coronary artery disease accounted for approximately 32% of all deaths globally in 2020, following an estimated 18.6 million fatalities in 2019, with annual global casualties projected to reach 23.6 million by 2030. In the United States, data from the National Health and Nutrition Examination Survey revealed a 48% disease prevalence among adults aged 20 and older between 2013 and 2016, with incidence climbing progressively alongside advancing age. Although standard coronary angiography provides direct visualization of intraluminal arterial narrowing, procedural hazards, substantial costs, and the requirement for specialized operators prevent its use as an indiscriminate screening mechanism. Meanwhile, routine non-invasive clinical testing such as electrocardiography and echocardiography often lacks sufficient diagnostic sensitivity and specificity, creating an operational gap that automated machine learning frameworks aim to bridge.

Data preparation within the study involved systematic curation of the primary repository before model training. The raw Z-Alizadeh Sani dataset comprises 55 recorded variables, consisting of 54 independent diagnostic indicators alongside one binary dependent target that records patient classification. Diagnostic attributes encompass patient symptoms, physical signs, electrocardiogram outputs, and laboratory blood tests. Automated inspection using Python software libraries confirmed the complete absence of missing data entries across all 303 clinical records. To prevent features with larger scales from disproportionately steering the models, the researchers normalized 21 continuous quantitative variables using a RobustScaler algorithm. Unlike standard scaling methods that divide centered values by standard deviation, this formulation calculates scaling by subtracting the feature median and dividing by the interquartile range between the 25th and 75th percentiles. Centering on the median and middle 50 percent of observations preserves the intrinsic distribution geometry and prevents non-normal measurements or extreme clinical outliers from distorting numerical representations.

Evaluation across baseline classification algorithms revealed significant performance disparities depending on architecture and balancing technique. Before applying ensemble feature selection and hyperparameter optimization, the standard unoptimized support vector machine yielded an accuracy of 77.00%. Among standalone traditional models, logistic regression achieved 87.77% accuracy under grid search and 83.48% under random search, demonstrating competitive performance that remained constrained by its inability to capture non-linear feature interactions. Random forest reached 83.86% accuracy with grid search and 86.77% with random search, whereas XGBoost attained 84.18% and 84.15% across the respective optimization strategies. In sharp contrast, configuring the support vector machine with CT-GAN generated tabular data yielded only 70.65% accuracy under grid search and 68.32% under random search. CT-GAN produced a lower specificity of 99.09% compared to the 100% specificity attained by ADASYN, which the authors attributed to generative training instability and mode collapse when attempting to model complex medical relationships from only 303 samples.

External cross-validation across three public cardiovascular databases further defined the generalizability boundaries of the methodology. On the Cleveland dataset containing 303 samples across 14 clinical features, the optimized support vector model achieved 77.20% accuracy, 72.59% sensitivity, 81.10% specificity, and an area under the curve of 0.8362. Testing on the Statlog repository of 270 subjects and 13 variables produced 76.30% accuracy, 72.50% sensitivity, 79.33% specificity, and an area under the curve of 0.7667. The lowest generalizability emerged on the Long Beach Veterans Administration cohort of 200 patients across 13 attributes, which recorded 66.00% accuracy, 73.75% sensitivity, an area under the curve of 0.6594, and a specificity of 43.27%. The authors noted that this pronounced drop in specificity highlights the difficulty of transferring diagnostic models across disparate healthcare environments with varying patient demographics and reduced feature counts, reaffirming the necessity of tailoring predictive algorithms to specific clinical settings.

Subject of Research: Machine learning for non-invasive coronary artery disease diagnosis

Article Title: Predicting coronary artery disease with ensemble-based feature selection and grid search-tuned SVM under class imbalance conditions

Article References: Pahlevani, A., Kashefi, A., Arslan, E., Kumeda Kussia, B., & Hassannataj Joloudari, J. (2025). Predicting coronary artery disease with ensemble-based feature selection and grid search-tuned SVM under class imbalance conditions. Discover Data, 3(1), Article 48. https://doi.org/10.1007/s44248-025-00088-6

Image Credits: AI Generated

DOI: 10.1007/s44248-025-00088-6

Keywords: Coronary Artery Disease, Machine Learning, Support Vector Machine, Feature Selection, ADASYN, Class Imbalance, Cardiovascular Risk Factors, Clinical Decision Support, SHAP Analysis, Nested Cross-Validation, Medical Informatics, Non-invasive Diagnosis, Ensemble Learning, Hyperparameter Optimization, Computational Medicine, Heart Disease Prediction

News Source: Teresa Odom. (November 16, 2026). Optimized Machine Learning Model Predicts Coronary Artery Disease with High Accuracy. Scienmag.

Tags: ADASYNCardiovascular risk factorsclass imbalanceClinical decision supportComputational MedicineCoronary artery diseaseensemble learningFeature SelectionHeart Disease Predictionhyperparameter optimizationMachine LearningMedical Informaticsnested cross-validationNon-invasive DiagnosisSHAP analysissupport vector machine
Share12Tweet7Share2ShareShareShare1

Related Posts

New Controller Stops Collisions and Cuts Angular Jerk in Omnidirectional Robots

New Controller Stops Collisions and Cuts Angular Jerk in Omnidirectional Robots

November 16, 2026
Re-Derived Borisov Model Connects Grain Boundary Free Energy Segregation and Diffusion

Re-Derived Borisov Model Connects Grain Boundary Free Energy Segregation and Diffusion

November 16, 2026

Closed-Loop Optimization Adds Traceable Structural Interpretability to 3D Gaussian Splatting

November 16, 2026

Cast Iron Yields Free-Standing Graphite Anodes for Lightweight Lithium-Ion Batteries

November 16, 2026

POPULAR NEWS

  • Longitudinal Molecular Profiling in ALK-Rearranged Lung Cancer: A Case Report

    Longitudinal Molecular Profiling in ALK-Rearranged Lung Cancer: A Case Report

    29 shares
    Share 12 Tweet 7
  • Periodontal Ligament Simulation Variations Do Not Alter Root Fracture Resistance In Vitro

    29 shares
    Share 12 Tweet 7
  • Museum Records Reveal 33,000 Non-Asilid Insect Specimens Collected by Jason Londt

    29 shares
    Share 12 Tweet 7
  • State Reconstruction Model Slashes Epidemic Infection Costs by 88 Percent

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Arabic Schizophrenia Quality of Life Questionnaire Shows Reliability in Lebanese Inpatients

Surgery Surpassed Botulinum Toxin for Poststroke Upper Extremity Spasticity in Trial

Case Report Documents Concurrent Fungal Endophthalmitis and Cytomegalovirus Retinitis

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 87 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.