Authors

• P. Naga Srinivasu, Department of Computer Science and Engineering, Prasad V. Potluri Siddhartha Institute of Technology, Vijayawada, India.
• Uddagiri Sirisha, Department of Computer Science and Engineering, Prasad V. Potluri Siddhartha Institute of Technology, Vijayawada, India.
• Kotte Sandeep, Department of Information Technology, Dhanekula Institute of Engineering & Technology, Vijayawada, India.
• S. Phani Praveen, Department of Computer Science and Engineering, Prasad V. Potluri Siddhartha Institute of Technology, Vijayawada, India.
• Lakshmana Phaneendra Maguluri, Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, India.
• Thulasi Bikku, Computer Science and Engineering, Amrita School of Computing Amaravati, Amrita Vishwa Vidyapeetham, Amaravati, India.

Background

Heart stroke is one of the leading causes of morbidity and mortality worldwide, making early detection and prevention critical in reducing its impact. Predictive models using machine learning (ML) and deep learning (DL) techniques have been extensively researched for heart stroke prediction. However, a significant barrier to the widespread adoption of these models in clinical settings is their interpretability. Clinicians require not only accurate predictions but also a clear understanding of the underlying factors contributing to those predictions, which traditional ML and DL models often fail to provide.

Challenges

The primary challenge addressed in this study is the development of a heart stroke prediction model that is both highly accurate and interpretable. While existing models may achieve high accuracy, their "black-box" nature limits their usability in clinical decision-making. Healthcare professionals need models that not only predict outcomes but also explain the reasoning behind these predictions in a way that is understandable and actionable.

Objective

The objective of this study is to create an interpretable and explainable AI-based model for heart stroke prediction that can be effectively utilized in clinical settings. The focus is on ensuring that the model not only provides high predictive accuracy but also offers transparency and insights into the factors influencing individual predictions.

Methodology

The research introduces an innovative approach that combines traditional ML techniques with explainable AI (XAI) methods to create a model that is both accurate and interpretable. The methodology involved several key steps:

Data Preparation: The study utilized the Stroke Prediction Dataset, which includes 11 attributes relevant to heart stroke risk. The data was preprocessed to address issues such as data imbalance and potential data leakage, ensuring the integrity and reliability of the model.

Model Development: An Artificial Neural Network (ANN) model was developed for heart stroke prediction. To enhance the model’s performance, techniques such as resampling, feature selection, and the prevention of data leakage were employed. The ANN model was trained and validated using the prepared dataset.

Explainable AI Techniques: To address the interpretability challenge, the study incorporated two key XAI techniques:

Permutation Importance: This method was used to provide global insights into which features were most significant in predicting heart stroke across the entire dataset.

Local Interpretable Model-agnostic Explanations (LIME): LIME was applied to generate local explanations for individual predictions, offering clinicians a clear understanding of why the model made a specific prediction for a particular patient.

Results

The implementation of the ANN model combined with XAI techniques yielded impressive results:

High Predictive Accuracy: The ANN model achieved a remarkable accuracy of 95% in predicting heart strokes. This high level of accuracy ensures that the model can reliably be used for early detection of heart stroke risks.

Global Interpretability: The use of permutation importance allowed the researchers to identify which attributes were most influential in the model’s decision-making process, providing a global view of feature significance. This helps clinicians understand the general factors contributing to heart stroke risk.

Local Interpretability: LIME provided instance-specific explanations, allowing clinicians to see why the model made a particular prediction for an individual patient. This feature is particularly valuable in a clinical setting, where understanding the reasoning behind a prediction is as important as the prediction itself.

Conclusion

This study successfully demonstrates that it is possible to create a heart stroke prediction model that is both highly accurate and interpretable, addressing a critical need in healthcare. By combining an ANN model with XAI techniques, the researchers developed a tool that not only predicts heart stroke risk with high accuracy but also provides clear and understandable explanations for its predictions. This dual focus on accuracy and interpretability makes the model a valuable asset in clinical decision-making.

Implications

The implications of this research are significant for the field of healthcare, particularly in the early detection and prevention of heart strokes. The interpretability of the model means that it can be more easily integrated into clinical workflows, empowering healthcare professionals with both predictive insights and a deeper understanding of the factors driving those predictions. This can lead to more informed decision-making, better patient outcomes, and increased trust in AI-driven healthcare solutions.

Future Work

Future research could explore the application of similar interpretable AI techniques to other areas of healthcare, such as the prediction of other chronic diseases or the development of personalized treatment plans. Additionally, further refinement of the model and expansion to larger and more diverse datasets could enhance its robustness and applicability across different patient populations. The ongoing integration of explainable AI in healthcare promises to bridge the gap between advanced predictive models and practical clinical application, ultimately leading to better health outcomes.