The Artificial Intelligence Extended Hybrid Model Based on Metabolic and Fertility Data for Early Diagnosing Polycystic Ovary Syndrome in Iranian Fertile and Infertile Women and Providing Healthy Lifestyle Recommendations

Zahra Arab  Taheri Zadeh; Valiollah Dabidi Roshan; Tayebeh Gharaei

doi:10.18502/ijdl.v26i1.21333

Zahra Arab Taheri Zadeh Department of Exercise Physiology, Faculty of Physical Education, University of Mazandaran, Babolsar, Iran
Valiollah Dabidi Roshan Department of Exercise Physiology, Faculty of Physical Education, University of Mazandaran, Babolsar, Iran
Tayebeh Gharaei Faculty of Nursing and Midwifery, Iran University of Medical Sciences, Tehran, Iran

DOI: https://doi.org/10.18502/ijdl.v26i1.21333

Keywords: Polycystic ovary syndrome diagnosis, Hybrid machine learning models, Healthy lifestyle, Regular physical activity

Abstract

Background: Polycystic ovary syndrome (PCOS) is associated with metabolic, hormonal, and genetic disorders. The lack of defined biomarkers makes diagnosis difficult. High-accuracy hybrid models enable early diagnosis. The aim of the present study is to train a hybrid model with metabolic and reproductive indicators for early diagnosis and provide healthy lifestyle strategies.

Methods: Data from 7000 fertile and infertile women and those without PCOS were processed, and then a dataset of 550 women was prepared, and 7, 10 and 15 subsets of important features were selected using random forest (RF) and were used to train hybrid models Voting classifier, LG, SVC, XGBoost.

Results: After selecting three groups of important features and training the models, the Voting classifier model could diagnose PCOS with an accuracy of over 95%. Anti-Mullerian (AMH) is considered an important diagnostic tool. In addition, sex hormones and markers such as fasting glucose, total cholesterol, high-density lipoprotein cholesterol, vitamin D3, and thyroid hormones can be used for early diagnosis of this syndrome.

Conclusion: It is possible to identify polycystic ovary syndrome using machine learning models without expensive highprecision tests, which will help doctors and clinicians make informed decisions and reduce harmful messages.