Different medical data mining approaches based prediction of ischemic stroke

Area of Science:

• Medical data mining
• Computational medicine
• Biostatistics

Background:

• Medical data mining, or knowledge discovery in medicine, extracts patterns from large datasets.
• Predicting ischemic stroke is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To evaluate and compare the predictive performance of different medical data mining approaches for ischemic stroke.
• To identify the most effective data mining model for ischemic stroke prediction.

Main Methods:

• Utilized a dataset of 80 patients and 112 healthy individuals with 17 predictors.
• Employed Support Vector Machine (SVM), Stochastic Gradient Boosting (SGB), and Penalized Logistic Regression (PLR) models.
• Applied 10-fold cross-validation and grid search for parameter optimization, evaluating models using accuracy, AUC, sensitivity, specificity, PPV, and NPV.

Main Results:

• Support Vector Machine (SVM) achieved an accuracy of 0.9789 and an Area Under the ROC Curve (AUC) of 0.9783.
• Stochastic Gradient Boosting (SGB) demonstrated comparable performance with an accuracy of 0.9737 and an AUC of 0.9757.
• Penalized Logistic Regression (PLR) showed lower predictive performance with an accuracy of 0.8947 and an AUC of 0.8953.

Conclusions:

• Support Vector Machine (SVM) exhibited the superior predictive performance for ischemic stroke classification among the evaluated models.
• Both SVM and SGB models demonstrate significant potential for accurately classifying ischemic stroke, offering valuable tools for clinical application.