A Heterogeneous Ensemble Learning Method For Neuroblastoma Survival Prediction

Area of Science:

• Pediatric Oncology
• Computational Biology
• Machine Learning

Background:

• Neuroblastoma is a significant pediatric cancer with high mortality rates.
• Accurate survival prediction is crucial for effective neuroblastoma treatment planning.

Purpose of the Study:

• To develop an advanced ensemble learning model for predicting neuroblastoma patient survival.
• To extract interpretable decision rules from the model to support clinical decision-making.

Main Methods:

• A heterogeneous ensemble learning approach integrating five base learners: decision tree, random forest, genetic algorithm-based support vector machine, extreme gradient boosting, and light gradient boosting machine.
• Heterogeneous feature selection to optimize subsets for each base learner, serving as prior knowledge.
• An area under the curve (AUC)-based ensemble mechanism for integrating base learners.

Main Results:

• The proposed method achieved high performance metrics: 91.64% accuracy, 91.14% recall, and 91.35% AUC.
• The ensemble model significantly outperformed mainstream machine learning methods.
• Interpretable decision rules with >0.900 accuracy were successfully extracted.

Conclusions:

• The developed heterogeneous ensemble learning method enhances neuroblastoma survival prediction accuracy.
• Extracted decision rules can improve the performance of clinical decision support systems.
• This approach holds potential for improving survival rates in pediatric neuroblastoma patients.