Prediction of polycystic ovary syndrome using machine learning models: Addressing class imbalance and high dimensionality.

backgroundPolycystic ovary syndrome (PCOS) is a hormonal disorder that affects fertility and long-term health issues such as metabolic syndrome. Class imbalance is prevalent in PCOS datasets and leads to bias toward the majority class, resulting in poor identification of the minority class. This study evaluated the impact of varying levels of class imbalance and high dimensionality on the performance of various machine learning (ML) models through a simulation study. MATERIALS AND

methodsThis study involved an imbalanced PCOS dataset consisting of 41 clinical and physical parameters of 541 patients gathered from 10 hospitals across Kerala, India. This study evaluated the performance of ML models, namely k-nearest neighbor, decision tree (DT), random forest (RF), support vector machine, and XGBoost, under varying imbalance ratios (IRs). Stratified resampling was used to generate multiple imbalanced scenarios. Model performance was assessed using accuracy, sensitivity, specificity, precision, and F1 score.

resultsFor IRs up to 3, RF outperformed other methods, based on performance measures. For IR >3, XGBoost emerged as the top-performing method, followed by DT. In the case of IR exceeding 5, the accuracy of the minority class fell below 50%. Based on simulation results, RF was used to identify significant variables.

conclusionFor original data (IR = 2.03), the RF model is a better model for predicting PCOS, with a sensitivity of 94% and a specificity of 81%. This study contributes to making informed decisions on selecting the most appropriate classification method based on a specific proportion of imbalance in healthcare datasets.