An efficient interpretable stacking ensemble model for lung cancer prognosis.

Area of Science:

• Oncology
• Machine Learning
• Bioinformatics

Background:

• Lung cancer is a leading cause of cancer mortality worldwide.
• Accurate prognosis is essential for effective lung cancer management and patient outcomes.
• Existing models often lack interpretability, hindering clinical trust and adoption.

Purpose of the Study:

• To develop an interpretable stacking ensemble model (SEM) for lung cancer prognosis prediction.
• To identify key risk factors influencing lung cancer prognosis.
• To compare the interpretability and performance of SEM against traditional machine learning models.

Main Methods:

• Utilized a Kaggle dataset comprising 1000 patients and 22 variables.
• Developed a stacking ensemble model (SEM) for classifying prognosis into Low, Medium, and High-risk categories.
• Employed bootstrap methods for evaluation and SHAP/LIME for model interpretability assessment.

Main Results:

• The SEM achieved high performance metrics: 98.90% accuracy, 98.70% precision, 98.85% F1 score, 98.77% sensitivity, 95.45% specificity, 94.56% Cohen's kappa, and 98.10% AUC.
• Demonstrated superior interpretability compared to Random Forest, Logistic Regression, Decision Tree, Gradient Boosting Machine, Extreme Gradient Boosting Machine, and Light Gradient Boosting Machine.
• Identified chronic lung cancer and genetic risk as significant factors in lung cancer prognosis.

Conclusions:

• The interpretable SEM offers a robust and reliable tool for lung cancer prognosis prediction.
• The model's interpretability enhances clinical trust and facilitates the identification of critical risk factors.
• Findings underscore the importance of chronic lung disease and genetic predisposition in lung cancer outcomes.