An Improved Prediction of Polycystic Ovary Syndrome Using SMOTE-based Oversampling and Stacking Classifier

PCOS is an endocrine disorder that affects female fertility in women who are of reproductive age or who have gradually passed it. This medical condition increases the risk of complex long-term issues. Missed periods, Irregular periods, or high light periods are some of the most typical signs of this condition. It can cause infertility, cardiovascular disease, type-2 diabetes, and ovulation failure. Early detection and therapy are known to lower the threat of developing PCOS. Our objective is to identify which classification model is best in the prediction of disease. We use fourteen machine learning algorithms to compare and get the best model. To eliminate class imbalance and improve classification performance, data were re-sampled using Synthetic Minority Over-Sampling Technique (SMOTE). We examined several model assumptions using assessment metrics like accuracy, precision, recall, and AUC score. To compare the outcomes, classifiers include Logistic Regression, Naïve Bayes, K-nearest Neighbor, Support Vector Machine, Multi-layer Perceptron, Decision Tree, Gradient Boosting, AdaBoost, CatBoost, LightGBM, XGBoost, Random Forest, Voting and Stacking using GridSearchCV technique has been used. The stacking classifiers achieved the highest 98% of accuracy in determining PCOS from clinical data of women patients and outperformed the previous works. Our research will aid the medical community in decreasing the miscarriage rate.