Ovarian cancer is often detected only in advanced stages. Early symptoms are either absent or very mild, which makes them difficult to notice. Patients usually fail to recognize these signs, and primary care doctors also face difficulty in separating them from other common conditions. This study looks into how machine learning, can help find ovarian cancer early through blood tests. CatBoost algorithm is used to construct a predictive model on a dataset containing 48 blood biomarkers. Key findings from this work include that the top 18 most important features being discovered by removing redundant features using the MRMR feature selection. The data underwent correlation analysis to know the interrelations between features of importance and target variables. CatBoost model with MRMR feature selection showed outstanding performance of accuracy of 97.7%, higher than other compared models. It also maintained strong precision, recall, and F1-score across both positive and negative cases. Training and validation curves showed consistent learning behavior, with no signs of overfitting. Based on the findings, it can be said that CatBoost is a good choice for identifying ovarian cancer early, which could be a big deal in future healthcare research. Further, research can be performed on bigger groups of patients and see how well the model works for different kinds of patients in the future.