Predicting Ovarian Cancer using Machine Learning: A Comparative Study

Ovarian cancer is one of the most lethal gynecological malignancies worldwide, frequently dubbed a "silent killer" due to its non-specific symptoms in early stages leading to poor 5-year survival rates. This study presents a comparative analysis evaluating classical statistical baselines, tree-based ensemble learning architectures, deep multi-layer perceptrons, and hybrid models in predicting ovarian malignancy using multi-parametric clinical features and serum biomarker profiles (CA125, HE4, CEA). Tree-based ensemble models, specifically XGBoost and LightGBM, significantly outperformed classical diagnostic indices (RMI and ROMA), achieving ROC-AUC scores of 0.96 and 0.95 with sensitivities exceeding 93%. Serum CA125 and HE4 levels were confirmed as the dominant predictive clinical features. The proposed machine learning framework serves as an automated early triage tool in clinical healthcare settings to assist oncologists in early risk stratification and reduce diagnostic delays.