An understandable PCOS prediction process using pre-trained and deep learning network model

Biochemical hyperandrogenism, anovulation, oligomenorrhea, and sometimes the existence of ovarian microcysts are the main characteristics of PCOS, a complicated illness. Such endocrinopathy results in symptoms like infertility, weight gain, acne, and hirsutism by preventing the formation of ovarian follicles. Healthcare data are handled by artificial intelligence (AI) which made substantial contributions to the fields of technology. To identify PCOS in healthy patients, we have thus shown an AI method utilizing multimodal Deep Learning (DL) classifiers. We made use of 541 individuals from an openly accessible dataset from Kerala, India. The conclusive multi-stack MobileNetV3 model with pre-trained CNN performed better than the other classifiers with 98.5%, 98%, 99%, and 98.5% for recall, accuracy, F1-score, and precision. Model identifications are provided intelligibly, interpretably, and consistently through AI techniques. To clarify tree-based classification algorithms, we have employed AI techniques and feature significance with mutual information. To help healthcare providers make decisions, this investigation aims to reliably identify PCOS among individuals while suggesting an automated diagnostic structural design with comprehensible ML capabilities.