Machine Learning Models for the Development of a Probabilistic Screening Tool for Polycystic Ovary Syndrome

Abstract

Background: Polycystic ovary syndrome (PCOS) is a common hormonal disorder in women of reproductive age that can lead to infertility and other long-term health problems. Early detection using simple, non-invasive tools is important to support timely intervention and improve outcomes. Objective: The study aimed to compare the performance of decision tree and naive Bayes models in predicting the likelihood of PCOS using non-invasive clinical features. Methodology: The study included 100 diagnosed cases of PCOS and 100 controls based on ultrasonographic findings. Clinical and lifestyle information was collected through a structured questionnaire. The models were evaluated using accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve. Five-fold cross-validation was used for validation. Results The decision tree model had high training accuracy but lower test accuracy, indicating overfitting. The naive Bayes model showed more consistent performance with 81 percent test accuracy and an F1 score of 0.81. Conclusion: The naive Bayes model shows promise as a simple, non-invasive screening tool for early identification of PCOS, particularly in primary care and low-resource settings.