Interpretable artificial intelligence (AI) for cervical cancer risk analysis leveraging stacking ensemble and expert...

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Cancer Research

Background:

• Cervical cancer prediction requires accurate and interpretable models.
• Traditional machine learning (ML) models often lack transparency.
• Explainable Artificial Intelligence (XAI) is crucial for clinical adoption.

Purpose of the Study:

• To develop an explainable ML system for cervical cancer prediction.
• To enhance predictive accuracy and model stability using hybrid feature selection.
• To integrate XAI techniques for transparent and trustworthy clinical decision support.

Main Methods:

• A hybrid feature selection combining correlation-based selection and recursive feature elimination.
• Ensemble modeling integrating random forest, extreme gradient boosting, and logistic regression.
• Integration of global and local XAI techniques for model interpretation.

Main Results:

• The ensemble model achieved 98% accuracy and 99.50% AUC, outperforming other models.
• Feature selection and data balancing significantly improved classification stability.
• XAI techniques and domain expert validation confirmed the practical relevance of key features.

Conclusions:

• Hybrid feature selection and ensemble learning significantly improve cervical cancer prediction.
• XAI integration enhances transparency, interpretability, and trustworthiness for clinical use.
• The developed system shows significant potential for clinical decision-making in cervical cancer detection.