Development of a Random Forest-Based Predictive Model for Polycystic Ovary Syndrome (PCOS) Using SHAP for Explanability

Abstract

Polycystic Ovary Syndrome (PCOS) is a lead major disorder and primary cause of infertility in women of reproductive age, affecting about 13% of this population globally with over 70% of cases remaining undiagnosed. Early diagnosis is yet challenging, particularly in low-resource settings where ultrasound imaging is inaccessible. This study focuses on leveraging Random Forest (RF) model for PCOS prediction using only clinical and biochemical data, enhanced with Shapley Additive Explanations (SHAP) for model interpretability. A publicly available Kaggle PCOS dataset from 541 women (177 PCOS-positive, 364 negative) across 10 hospitals in Kerala, India, was utilised. A leakage-free preprocessing pipeline applied median and mode imputation before data splitting.