Improved Prediction of Ovarian Cancer Using Ensemble Classifier and Shaply Explainable AI.

Area of Science:

• Oncology
• Artificial Intelligence in Medicine
• Biostatistics

Background:

• Ovarian cancer, a 'silent killer,' presents challenges in early detection due to subtle initial symptoms.
• Late-stage diagnosis significantly reduces treatment efficacy and survival rates for ovarian cancer.
• Routine screenings like pelvic exams, ultrasounds, and biomarker blood tests are vital for early ovarian cancer detection.

Purpose of the Study:

• To develop a highly accurate predictive model for ovarian cancer detection using a comprehensive dataset.
• To enhance the reliability and accuracy of ovarian cancer prediction through advanced machine learning techniques.
• To validate the model's performance against existing state-of-the-art methods and ensure interpretability.

Main Methods:

• Utilized the Soochow University ovarian cancer dataset with 50 distinct features.
• Developed a stacked ensemble model integrating bagging and boosting classifiers for enhanced predictive power.
• Employed SHAPly (SHAP) for explainable artificial intelligence to elucidate model predictions.

Main Results:

• The proposed stacked ensemble model achieved a record accuracy of 96.87% on the dataset using all 50 features.
• The model demonstrated superior performance compared to other cutting-edge models in ovarian cancer prediction.
• SHAPly analysis provided insights into the factors driving the model's accurate predictions.

Conclusions:

• The developed stacked ensemble model represents a significant advancement in accurate and reliable ovarian cancer detection.
• High accuracy in early detection can lead to improved treatment strategies and better patient survival rates.
• Explainable AI methods like SHAPly are crucial for understanding and trusting AI-driven diagnostic tools in oncology.