<p>Polycystic ovarian syndrome (PCOS), a disorder frequently linked to increased levels of testosterone and other androgens, is especially harmful to women of reproductive age. It is commonly acknowledged that PCOS has a significant role in ovulatory failure and an elevated risk of miscarriage. According to recent epidemiological data, this illness affects about 31.3% of Asian women. Despite its pervasive influence, further research is urgently needed to create efficient treatment plans for the potentially fatal side effects of PCOS. Several machine learning (ML) techniques have been employed in earlier research to classify PCOS automatically. However, the trustworthiness of these conventional ML techniques for precise PCOS detection is limited because they usually rely on manually designed features and have subpar performance. This work aims to enhance PCOS diagnosis and prediction by improving predictive accuracy and reliability by employing sophisticated deep learning approaches that automate feature extraction using PCOSVisionNet (PCOSDVNet). The lightweight deep learning approach are introduced by the suggested framework. It incorporates the residual connection to overcome class imbalance challenges and guarantee robust performance. The models’ performance metrics were outstanding. The ROC-AUC values were also statistically compared between models using the statistical test. The PCOSDVNet outperformed the other model assessed in terms of accuracy, precision, recall, ROC-AUC score, parameter efficiency, training duration, and statistical significance, among other metrics. A comparison with current cutting-edge PCOS diagnosis techniques validates the suggested PCOSDVNet-based model’s increased efficacy. These findings imply that this approach may help with early PCOS detection and lower miscarriage risk, which would ultimately lead to better reproductive health outcomes.</p>