Improving stroke risk prediction by integrating XGBoost, optimized principal component analysis, and explainable...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Stroke is a leading cause of disability and mortality globally.
• Existing stroke risk prediction models require improved efficiency and interpretability.
• Cerebrovascular diseases represent a significant public health concern.

Purpose of the Study:

• To enhance stroke risk prediction models by integrating advanced machine learning and dimensionality reduction techniques.
• To improve the efficiency and interpretability of stroke risk assessment.
• To introduce explainable artificial intelligence (XAI) into the principal component analysis (PCA) process for better risk factor understanding.

Main Methods:

• Utilized XGBoost and optimized principal component analysis (PCA) for data structuring and processing.
• Integrated explainable artificial intelligence (XAI) into PCA for enhanced transparency and interpretation.
• Employed OpenMP parallelization to accelerate processing speed.
• Validated the approach on two distinct datasets.

Main Results:

• Achieved high prediction accuracy of 95% and 98% on tested datasets.
• Demonstrated excellent model generalizability and reliability with cross-validation yielding an average of 0.99.
• Obtained high Matthew's correlation coefficient (MCC) of 0.96 and Cohen's Kappa (CK) of 0.96.
• Increased processing speed threefold through OpenMP parallelization.

Conclusions:

• The proposed method offers an innovative and reliable approach to stroke risk prediction.
• Integration of XAI with PCA enhances the interpretability of risk factors for medical professionals.
• The threefold increase in processing speed makes the model practical for real-world healthcare applications.
• This approach has the potential to significantly improve forecasting systems in the healthcare industry.