AI meets Reproductive Health: Early Diagnosis of PCOS using Machine Learning
TL;DR
The findings aid in describing the significance of the machine learning methodology to be supplemented by the analytical methods to produce useful clinical data, enhance the quality of diagnosis, and promote the initial PCOS detection, thus contributing to the general achievement of AI in the field of prophylaxis and personalised medicine.
Abstract
Polycystic Ovary Syndrome (PCOS) is a common complex hormonal condition that disrupts the balance in metabolism, fertility, and dermatological health, especially among women of reproductive age. It's mixed, and superimposing clinical analysis often tends to slow down the correct diagnosis. However, in the recent past, the convergence of data science and healthcare has made it possible to analyse large-scale patient data to gain a deeper understanding of complex diseases such as PCOS. The suggested methodology implies the use of a multi-dimensional database comprising physiological measurements, hormone concentration, lifestyle factors, and clinical indicators to find trends that can be related to the condition. A data-driven systematic method is used to examine statistical relations and hidden patterns. Pre-processing of data (dealing with missing values, outlier detection, normalising values and categorical encoding) is done, followed by the Exploratory Data Analysis (EDA), which reveals the major associations and visual representations of the distributions of variables. The feature engineering adds new features like hormone ratios, BMI groups and waist to hip measurements to enhance model performance. Improved oversampling methods, namely ADASYN and Borderline-SMOTE, reduce the imbalance between classes, and state-of-the-art models of machine learning, such as Logistic Regression, Random Forest, XGBoost, CatBoost, and ensemble voting classifiers, are applied to the PCOS classification. Clinical relevance is statistically tested (Chi-square, t-tests, ANOVA), and the explainable AI using SHAP is used to complement model interpretability. The findings aid in describing the significance of the machine learning methodology to be supplemented by the analytical methods to produce useful clinical data, enhance the quality of diagnosis, and promote the initial PCOS detection, thus contributing to the general achievement of AI in the field of prophylaxis and personalised medicine.