Hybrid Machine Learning Framework for Clinical Diagnosis of Polycystic Ovary Syndrome

Polycystic Ovary Syndrome (PCOS) is a common hormonal disorder in women, and it is difficult to diagnose at an early stage due to varying symptoms and limitations of traditional diagnostic methods. Early detection of PCOS is important to prevent long-term health complications. In this study, a hybrid machine learning model is proposed for PCOS detection using clinical and hormonal data. A dataset containing 541 patient records was used for analysis. Missing values were imputed using K-Nearest Neighbour (KNN), and the most relevant features were selected using Mutual Information. To address class imbalance, SMOTE was applied to the training data. Individual machine learning models were first evaluated, and based on their performance, a hybrid model was developed using a weighted soft-voting approach that combines Gaussian Naïve Bayes, Logistic Regression, and Random Forest. The experimental results suggest that the hybrid model strikes a better balance between accuracy, precision, and recall than any single model on its own. This makes it a more trustworthy approach for predicting PCOS. This project marks just the first phase of research, laying the groundwork for future studies that will combine these findings with ultrasound image analysis.