XAI-driven CatBoost multi-layer perceptron neural network for analyzing breast cancer

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Early breast cancer diagnosis is critical for effective treatment outcomes and women's health.
• Interpretable models are needed to understand the factors driving diagnostic predictions.
• Current breast cancer identification techniques can be enhanced with advanced machine learning.

Purpose of the Study:

• To propose a novel CatBoost+MLP model for breast cancer data analysis.
• To enhance the interpretability of breast cancer diagnostic predictions using explainable AI.
• To improve feature identification and decision-making transparency in cancer diagnosis.

Main Methods:

• Utilized a hybrid CatBoost classification and Multi-Layer Perceptron (MLP) neural network model.
• Employed explainable artificial intelligence (XAI) techniques, specifically Shapley Additive Explanations (SHAP), for interpretability.
• Performed feature engineering using Analysis of Variance (ANOVA) to identify significant predictive features.

Main Results:

• The CatBoost+MLP model demonstrated enhanced interpretability in breast cancer diagnosis.
• SHAP values were used to analyze feature significance, providing insights into model decisions.
• Performance metrics showed the efficacy of the proposed model compared to contemporary methods.

Conclusions:

• The developed CatBoost+MLP model offers a more interpretable approach to breast cancer diagnosis.
• Leveraging CatBoost for feature identification and MLP for prediction improves diagnostic transparency.
• This approach contributes to more trustworthy AI-driven medical decision-making in oncology.