Polycystic Ovary Syndrome (PCOS) is a complicated Endocrine-based disorder, which globally affects nearly 8–13% of women at their reproductive ages. Specifically, PCOS causes significant psychological and long term health issues, since it is closely linked with metabolic complications, obesity, Type-2 diabetes and dyslipidemia. Although, PCOS is a common disorder, its exact cause remains unclear, which requires efficient detection and treatment at the early-stages of the disease. The existing literature on PCOS prediction are primarily focusing on identification of the disease, whereas less attention is given towards optimization strategies, which play a significant role in enhancing the interpretability as well as predictability of the models. To solve these issues, this research study introduces a novel Optimized CatBoost classifier-based PCOS prediction system using Feature selection and Threshold tuning strategies, which can efficiently detect PCOS from prodromal symptoms. The evaluations conducted on Benchmark datasets demonstrate the enhanced results of proposed model when compared to the baseline techniques in terms of several performance metrics such as Precision, Accuracy, Recall, ROC-AUC and F1-Scores.