PCOScare: Conventional Machine Learning Classifiers for Diagnosing and Prevention

Abstract

PCOS is a common endocrine disorder that affects women of reproductive age. Timely identification and diagnosis of PCOS are essential for the successful treatment and prevention of related health complications. ML techniques have shown promise in automating PCOS diagnosis using various clinical and biochemical features. Though, the presentation of these mockups seriously relies on the selection of relevant features, as including irrelevant or redundant features can lead to overfitting and decreased generalization performance. In this study, we propose an optimized feature selection approach for PCOS detection using ML algorithms. We first compile a comprehensive dataset comprising clinical and biochemical features commonly associated with PCOS, including hormone levels, menstrual irregularities, and anthropometric measurements. For the purpose of handling missing data and scaling features properly, feature preprocessing techniques like normalization and imputation are used. We next investigate several feature selection strategies, such as filter, wrapper, and embedding approaches, to find the most relevant characteristics for PCOS identification. We use cross-validation and presentation needles with accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC-ROC) in a methodical review approach to maximize feature selection. The presentation of respectively feature assortment method is compared to assess its effectiveness in identifying discriminative features for PCOS diagnosis. Furthermore, we investigate the impact of feature selection on different ML algorithms, including support vector classifier, random forests, and Xg-boosting classifiers. Our results demonstrate that feature selection significantly improves the performance of PCOS detection models by reducing dimensionality and focusing on the most relevant features. Moreover, we identify key features that contribute most to the discriminative power of the models, providing insights into the underlying characteristics of PCOS. The optimized feature selection approach proposed in this study offers a promising strategy for developing accurate and interpretable ML models for PCOS detection, ultimately aiding clinicians in early diagnosis and personalized treatment planning.