Enhancing PCOS diagnosis accuracy with modified Giza Pyramids Optimizer, SLOGPO, and PCOSNet: advanced classification and optimization techniques

Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting women during their reproductive years, characterized by a range of hormonal imbalances and reproductive dysfunctions. Accurate diagnosis of PCOS is crucial, as it serves as the foundation for timely interventions and personalized management strategies. In this context, this paper presents a comprehensive evaluation of two advanced classification models: the Stacked Learning Optimized by Giza Pyramids Optimizer (SLOGPO) and PCOSNet, a model leveraging 1-Dimensional Convolutional Neural Networks (CNN) integrated with a BORUTA-based feature selection framework. The study employs three distinct datasets, each containing a variety of physical and clinical parameters. The SLOGPO model, which combines stacked learning, the Improved Giza Pyramids Optimizer (IGPO), and BORUTA for feature selection, demonstrated strong generalization capabilities, achieving accuracies of 96% on Dataset I, 96% on Dataset II, and 95% on Dataset III. In contrast, the PCOSNet model, based on a 1D CNN architecture with BORUTA, achieved accuracies of 94% on Dataset I, 93% on Dataset II, and 84% on Dataset III. These results underscore the superior generalization ability of the SLOGPO model across varied datasets, while the performance of PCOSNet indicates areas for further improvement, particularly in handling diverse datasets. This study provides valuable insights into the impact of feature selection techniques and dataset variability on the performance of PCOS detection models. The findings emphasize the importance of evaluating models across multiple datasets to assess their generalizability and robustness, laying the groundwork for future research in PCOS detection and related medical applications.