Polycystic Ovary Syndrome (PCOS) is a frequent endocrine illness in women of reproductive age, often leading to infertility, metabolic disorder, and cardiovascular disease. Early and appropriate diagnosis is critical for effective management, but routine diagnostic methods by clinical, biochemical, and ultrasonographic features are time-consuming and prone to variability. Machine learning (ML) presents a promising solution to enhance PCOS detection by identifying intricate patterns in patient data. This research utilizes XGBoost and CatBoost, two efficient gradient-boosting models, to create predictive models for PCOS diagnosis. The models are trained using a dataset of clinical, biochemical, and lifestyle features. Feature selection methods are utilized to determine the most important predictors, maximizing model performance. The models are tested using accuracy, precision, recall, and F1-score to ensure their reliability in clinical use.Experimental results demonstrate that XGBoost and CatBoost are better than classical models of classification with greater accuracy and stability of prediction in PCOS.