An Interpretable Approach with Explainable AI for Heart Stroke Prediction.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Artificial Intelligence for Disease Prediction

Background:

• Heart strokes pose a significant global health challenge.
• Existing machine learning (ML) models for stroke prediction often lack clinical interpretability.
• Healthcare professionals require understandable AI tools for critical decision-making.

Purpose of the Study:

• To develop an effective and interpretable heart stroke prediction model using explainable AI (XAI).
• To address the gap between complex ML models and their clinical applicability.
• To provide a reliable and understandable tool for healthcare practitioners.

Main Methods:

• Utilized the Stroke Prediction Dataset with 11 attributes.
• Implemented data preprocessing techniques: resampling, data leakage prevention, and feature selection.
• Employed explainable AI methods: permutation importance and LIME for model interpretability.

Main Results:

• Achieved an outstanding accuracy rate of 95% for heart stroke prediction.
• Permutation importance provided global feature insights.
• LIME offered local, instance-specific explanations, enhancing Artificial Neural Network (ANN) model comprehension.

Conclusions:

• The developed ANN model offers a reliable and interpretable solution for heart stroke prediction.
• Explainable AI techniques significantly improve the clinical utility of predictive models.
• This approach can enhance healthcare decision-making and patient outcomes in cardiovascular health.