Early Detection of Polycystic Ovary Syndrome: A Systematic Evaluation of Machine Learning Models Using Clinical Parameters

Polycystic Ovary Syndrome (PCOS) is a common but complex endocrine and metabolic disease that affects women of childbearing age. In severe cases, it can lead to infertility, recurrent miscarriage, pregnancy complications, and long-term cardiovascular and metabolic diseases. It is reported that the prevalence of PCOS in Asia can reach 31.3%, and more than 60% of women fail to be diagnosed and intervened in time due to the lack of obvious symptoms in the early stage of the disease. The pathogenesis of PCOS is closely related to androgen excess, ovulatory disorders, and abnormal follicular development. At present, it is urgent to build an efficient and explanatory prediction model to assist early screening and individualized intervention. This study aims to construct a PCOS prediction method with good explanatory power and practical feasibility based on clinical phenotypes and biochemical indicators. The study used a dataset of 41 features of 541 patients from multiple hospitals in southern India, and screened out 15 core indicators with strong medical relevance in combination with literature. Systematic data preprocessing and visualization analysis were carried out, and the performance of 13 machine learning models for binary classification tasks was evaluated. The results showed that the support vector machine (SVM) binary classification model had the highest prediction accuracy, reaching 86.76%, followed by the naive Bayes and custom kernel SVM models, both with accuracy rates exceeding 85%. The study not only provides an implementable path for early identification of PCOS and stratification of risk groups, but also provides an important reference for the construction of gynecological auxiliary diagnostic tools with strong interpretability and scalability in the future.