OptiSelect and EnShap: Integrating machine learning and game theory for ischemic stroke prediction

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Neuroscience

Background:

• Ischemic stroke prediction remains a critical challenge in healthcare.
• Machine learning models offer potential for improving diagnostic accuracy.
• Integrating game theory can enhance feature importance analysis in predictive models.

Purpose of the Study:

• To investigate the application of the Shapley value from game theory for predictive ischemic brain stroke analysis.
• To evaluate the performance of various machine learning models in identifying key stroke predictors.
• To leverage ensemble machine learning techniques for enhanced prediction accuracy.

Main Methods:

• Feature identification using preference algorithms across multiple machine learning models (logistic regression, KNN, decision trees, SVMs, neural networks).
• Evaluation of top 3, 4, and 5 features for predictive performance.
• Shapley value application to rank models based on their top four features.
• Ensemble methods applied to top-ranked models for final prediction.

Main Results:

• Identification of the most impactful features for ischemic stroke prediction.
• Shapley value analysis provided a robust method for ranking machine learning models.
• Ensemble models utilizing top-ranked methods achieved a high accuracy of 92.39%.
• The combined approach significantly outperformed individual models.

Conclusions:

• Game theory, specifically the Shapley value, is a valuable tool for feature selection and model interpretation in medical diagnostics.
• Ensemble machine learning methods significantly boost predictive accuracy in ischemic stroke analysis.
• The study demonstrates a powerful synergy between game theory and machine learning for clinical decision support.