Prediction of Polycystic Ovary Syndrome

Abstract Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting women of reproductive age, characterized by hormonal imbalance, metabolic irregularities, and reproductive complications. Early and accurate detection remains challenging due to heterogeneous symptoms and reliance on conventional diagnostic procedures, which are often time-consuming and subject to variability in clinical interpretation. Traditional methods struggle to effectively integrate multidimensional clinical and lifestyle data, limiting predictive reliability. To overcome these limitations, a data-driven predictive system is developed that incorporates comprehensive preprocessing, feature selection using Principal Component Analysis, and balanced sampling through synthetic data generation. Multiple classification models, including machine learning, deep learning, and ensemble techniques, are trained and evaluated to identify the most effective approach. The final system employs a soft voting ensemble model combined with SHAP-based explainability to provide interpretable predictions. Experimental evaluation demonstrates strong performance, with the ensemble model achieving an accuracy of 86.23%, precision of 82.64%, recall of 91.74%, and F1 score of 86.96%, indicating improved detection capability compared to individual models. The integration of explainable insights further enhances transparency and trust in predictions. Overall, the proposed system offers an efficient, accurate, and interpretable solution for early PCOS prediction. Keywords: Polycystic Ovary Syndrome (PCOS), Machine Learning, Ensemble Learning, Explainable Artificial Intelligence (XAI), Predictive Modeling, Healthcare Analytics. 1. Introduction Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting women of reproductive age, characterized by hormonal imbalance, metabolic irregularities, and reproductive complications. The condition is associated with long-term health risks such as infertility, insulin resistance, and cardiovascular diseases, making early detection essential for effective management. With the increasing availability of clinical and biomedical data, there has been a growing interest in utilizing intelligent computational approaches to support diagnosis and decision-making. Recent advancements highlight the potential of data-driven models to improve predictive accuracy and enable early intervention in PCOS detection 1, 2. These approaches provide an opportunity to transform traditional diagnostic practices into more efficient and scalable systems 3. Despite these advancements, several challenges persist in existing approaches. Conventional diagnostic methods rely heavily on clinical expertise and fragmented analysis, often leading to inconsistencies and delayed detection. Many computational systems fail to effectively handle heterogeneous data, class imbalance, and redundant features, which negatively impact predictive performance 4, 5. Furthermore, limited transparency in decision-making processes reduces trust and interpretability in healthcare applications 6. Although prior studies have explored predictive modeling using clinical and electronic health record data, gaps remain in achieving a balance between accuracy, robustness, and explainability 7, 8. These limitations highlight the need for more integrated and reliable frameworks capable of addressing real-world clinical complexities 9. The primary objective is to develop a comprehensive predictive framework that enhances early identification of PCOS while ensuring interpretability and consistency. The focus is on integrating structured clinical data into a unified system capable of analyzing multiple factors simultaneously. The contributions include the design of a systematic data processing workflow, incorporation of advanced learning strategies for improved predictive capability, and integration of interpretable mechanisms to provide meaningful insights into model decisions. Emphasis is placed on creating a scalable and user-oriented solution that bridges the gap between data availability and actionable healthcare insights 10. The significance of this work lies in its potential to support timely diagnosis and improve clinical decision-making. By enabling accurate and interpretable predictions, the system can assist healthcare professionals in identifying high-risk individuals at an early stage, thereby reducing complications and improving patient outcomes. Additionally, the integration of user-friendly interfaces and explainable outputs enhances accessibility and trust in automated systems. This approach contributes to the advancement of intelligent healthcare solutions by combining predictive performance with transparency, ultimately supporting more efficient, reliable, and patient-centric diagnostic processes. 2. Related Work Recent advancements in predictive modeling for Polycystic Ovary Syndrome (PCOS) have demonstrated the growing importance of data-driven approaches in healthcare. A comprehensive review by Memon et al. highlights the effectiveness of machine learning models in improving diagnostic accuracy and management strategies, while also emphasizing the need for standardized evaluation frameworks 11. Similarly, Velvizhi and Kiruthiknivas proposed a risk evaluation system that leverages advanced learning techniques to identify PCOS at early stages, showcasing improved predictive capabilities but limited scalability across diverse populations 12. Ghaderzadeh et al. further explored the role of artificial intelligence in PCOS diagnosis, noting its potential in handling complex clinical data while identifying challenges related to data heterogeneity and model interpretability 13. Further studies have focused on enhancing predictive performance using hybrid and deep learning approaches. Govindharajan et al. presented an extensive review of innovative AI-driven models, highlighting improvements in accuracy but also pointing out issues related to computational complexity and lack of explainability 14. Jain et al. emphasized the importance of explainable artificial intelligence, demonstrating how interpretability can enhance trust in clinical applications, although practical integration remains limited 15. Zigarelli et al. introduced machine-aided self-diagnostic models that enable early screening; however, these approaches often rely on limited datasets and may not generalize effectively 16. Narni et al. developed a preliminary screening tool using non-invasive features, which improves accessibility but may compromise predictive precision due to reduced feature richness 17. In addition to structured clinical data, some studies have explored alternative data modalities. Karthik et al. utilized image-based analysis for PCOS prediction, demonstrating the potential of deep learning in medical imaging but facing challenges related to data availability and preprocessing requirements 18. Lim et al. employed physiological signal analysis, such as pulse wave data, to classify PCOS cases, highlighting innovative feature sources but requiring specialized data acquisition methods 19. Aggarwal and Pandey analyzed traditional classification techniques, confirming their effectiveness while noting limitations in handling complex, non-linear relationships within medical datasets 20. Despite significant progress, several gaps remain in existing approaches. Many studies focus primarily on improving predictive accuracy without adequately addressing interpretability, data imbalance, and integration of heterogeneous features. Additionally, limited attention has been given to combining multiple modeling paradigms within a unified and deployable framework. The lack of user-centric systems that integrate prediction, explanation, and accessibility further restricts real-world applicability. To address these challenges, the current work focuses on developing a comprehensive and interpretable predictive framework that integrates diverse clinical features, ensures balanced learning, and provides transparent decision support. By combining accuracy with explainability and practical deployment considerations, the approach aims to bridge existing gaps and contribute toward more reliable and accessible PCOS detection systems. 3. Materials And Methods The proposed system aims to develop an accurate and interpretable predictive framework for early detection of Polycystic Ovary Syndrome using structured clinical data. The workflow begins with the integration of multiple datasets followed by rigorous preprocessing, including removal of redundant attributes, handling of missing values, and normalization using StandardScaler. Feature engineering is performed through Principal Component Analysis to identify high-impact attributes, from which a refined subset of significant features is selected to enhance model efficiency. To address class imbalance, Synthetic Minority Over-sampling Technique is applied, ensuring balanced learning. Multiple classification models, including traditional machine learning 21, deep learning architectures, and ensemble techniques, are trained and evaluated. A hybrid ensemble model combining stacking and soft voting mechanisms is employed to improve predictive robustness. Additionally, SHAP-based explainability is integrated to provide interpretable insights into model predictions. The system is expected to deliver high predictive accuracy along with transparent decision support, facilitating reliable and early identification of PCOS. Fig. 1. System Architecture Fig. 1 illustrates the overall system architecture, comprising presentation, application, machine learning, and deployment layers. User inputs are collected through the frontend and processed in the backend, where data validation, prediction handling, and result generation occur. The machine learning 22 pipeline performs preprocessi