One of the main causes of mortality for women is ovarian cancer, which can only be treated when it has progressed. Early discovery, however, can enhance patient survival and prognosis. The aim of this study is to assess how well different machine learning (ML) models perform in the early detection of ovarian cancer using clinical data and biomarker analysis. Clinical and biochemical data from 349 patients were used, along with general drug indicators like mean granulocyte ratio, alanine aminotransferase, and calcium levels, as well as significant biomarkers like carbohydrate antigen 125, carbohydrate antigen 19-9, carcinoembryonic antigen, and human epididymal protein 4. Eight ML models were evaluated using GridSearchCV for hyperparameter optimization: RandomForestClassifier, Support Vector Machine (SVM), Logistic Regression, GradientBoostingClassifier, LightGBM classifier, and custom implementation of LightGBM with Gradient Boosting. Among them, XGBoost and GradientBoostingClassifier show excellent performance, achieving the highest accuracy and power in ability to differentiate between situations that are benign and those that are malignant. In order to provide insights into the most significant biomarkers for each dataset, the study also incorporates SHAP for feature significance analysis. While tumor markers emphasized HE4 and CA125 as crucial, blood routine test datasets indicated NEU and AGE as essential aspects. The combined dataset reaffirmed the significance of HE4, CA125, and NEU, whereas the general chemistry data focused on AGE and ALB. This study shows how XAI improves transparency in medical predictions and emphasizes the significance of choosing the suitable ML model based on the dataset’s properties and performance measures. These results aid in the creation of more precise and understandable models for the detection of ovarian cancer.