A Novel Machine Learning Model for Predicting Natural Conception Using Non-Laboratory-Based Data.

This study aimed to predict the likelihood of natural conception among couples by using a machine learning (ML) approach based on sociodemographic and sexual health data. This marks a novel, non-invasive methodology for fertility prediction. This prospective study included 197 couples divided into two groups: 98 fertile couples (Group 1) who achieved natural conception within one year, and 99 infertile couples (Group 2) who were unable to conceive despite regular unprotected intercourse. Data were collected using a structured form capturing 63 variables from both partners. Using the Permutation Feature Importance method, 25 key predictors were selected. The variables included BMI, age, menstrual cycle characteristics, and varicocele presence. Five ML models were developed and their performance was evaluated using metrics such as accuracy, sensitivity, and specificity. The XGB Classifier showed the highest performance among the models tested with an accuracy of 62.5% and a ROC-AUC of 0.580, indicating limited predictive capacity. The selected predictors encompassed a balance of medical, lifestyle, and reproductive factors for both partners, emphasizing the couple-based approach. Key factors included BMI, caffeine consumption, history of endometriosis, and exposure to chemical agents or heat. This study assessed the use of ML to predict natural conception using sociodemographic and health data The key predictors identified emphasize the importance of couple-based and lifestyle factors in predicting natural conception. However, the predictive capacity of the models was limited, highlighting the need for future studies with larger datasets and expanded predictors to improve accuracy and facilitate AI integration into fertility assessment.