Optimized ML Framework for PCOS Diagnosis Using PSO-GA and Ensemble Learning

The symptoms of Polycystic Ovary Syndrome (PCOS) are varied as it is a complex endocrine disorder affecting women of reproductive age. Due to these drawbacks, traditional diagnostic methods rely on clinical evaluation, biochemical tests and ultrasound imaging being at times time consuming and inconsistent. Machine learning (ML) offers itself as a possible way of automating the diagnosis of PCOS through large scale clinical datasets as an approach of searching for patterns and increasing predictive accuracy. The optimized ML framework based on the data pre-processing, Mutual Information based feature selection and the KMeans-ENN based enhancement to overcome imbalance to achieve higher predictive accuracy of the model is proposed. Hyperparameter tuning is performed using proposed Particle Swarm Optimization-Genetic Algorithm (PSOGA) algorithm. Ensemble Learning comprises of the models like Random Forest, Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Gradient Boosting (GB), and Logistic Regression (LR). Before and after tuning, accuracy, precision, recall, and F1-score is assessed on the performance of the model. Results show that the hyperparameter tuning can significantly improve the classification performance, being in line with this case the best combination of precision and recall is attained by Logistic Regression and the best overall accuracy is by Random Forest.