Ovarian cancer is one of the most aggressive neoplasms worldwide, representing the eighth leading cause of cancer death in women, having a survival rate ranging from 90% for early diagnosis to 13.4% for advanced stages. This study aims to propose a reliable and accurate diagnostic support system via the development of Machine Learning models - Random Forest, Support Vector Machine, Decision Tree, Simple Artificial Neural Network and XGBoost - based exclusively on routine clinical and hematological parameters, characterised by low cost and easy availability. The study used a unified dataset of 933 samples and 48 common features, previously preprocessed using a hybrid imputation technique to handle missing values, improving performance compared to the reference study. The core of this work lies in the implementation of the Stacking Ensemble model with a Logistic Regression-based Meta Learner, which enhances the performance achieved by the most promising Base Learner. Experimental results demonstrate that the Stacking Ensemble outperforms the best individual models, achieving an accuracy of 0.9840 and a specificity of 1.0000. This evidence confirms the potential of Artificial Intelligence as a concrete support tool to improve the reliability of clinical decisions and early diagnosis.
