Improved PCOS Prediction Using SMOTE and Stacking Ensemble Classifiers
Abstract
Polycystic ovary syndrome refers to a condition that is quite common. What is more, it is found in women of reproductive age. Usually, this syndrome comes along with fertility problems. In addition, it is often associated with hormonal imbalance and plenty of metabolic disorders as well. It's challenging to diagnose the disease at early phase because of range of symptoms present and overlapping with other pathologies. Hence, the study presents to use ML-based techniques for predicting chances of having PCOS as per Kaggle dataset containing 541 patients' records with 44 features. The dataset undergoes preprocessing, followed by the imputation and normalization of missing values. The SMOTE technique is utilized to balance the classes in the data set. The most significant predictors are identified by random forest feature importance. After that, the RandomizedSearchCV algorithm is employed to tune the classifiers such as RF, SVM, XGBoost, and LightGBM and develop a stacking ensemble model using a Logit. The suggested method attains an accuracy of 93.58% and an ROC-AUC of 0.959. SHAP analysis was also used to identify the impact of different features on the model outputs.