Developing a model for prediction of cumulative conception with artificial intelligence approach

Cumulative conception refers to a woman's probability of becoming pregnant within a specific timeframe, influenced by various factors including social, cultural, economic, environmental, behavioral, biological, psychological, and genetic factors. It serves as an important indicator for women to consider when planning fertility or deciding on infertility treatment. Knowing which couples are at risk for infertility or subfertility is crucial in determining the need for preconceptional evaluation. Therefore, the development of non- invasive, affordable, and effective prediction models to explore the likelihood of achieving pregnancy is imperative. In a prospective thesis study conducted for this purpose, a registration form consisting of questions determined by a specialist physician was completed by 300 participants, and pregnancy follow-up was conducted every two months for up to a maximum of 12 months. Data obtained were used to identify variables with risk/chance factors through five different feature selection methods, and five different prediction models were developed using machine learning methods based on these variables. Among these models, the most successful one was achieved with the CatBoost algorithm (Accuracy:0.83, Sensitivity:0.90, Specificity:0.76 and ROC-AUC:0.83). Prediction results obtained according to the CatBoost algorithm were transformed into probability values, providing probability values for each class (pregnant/not pregnant) in the model output. These probability values were then divided into three classes "low," "medium," and "high" based on cutoff points determined by the specialist physician. A successful decision support model has been developed to assist physicians in providing an initial estimation of pregnancy likelihood as "low," "medium," or "high" solely based on registration form data during the preconception stage. With this developed decision support system, when a specialist physician enters candidate information into the system, they can obtain a preliminary prediction regarding the probability of pregnancy, along with using examination information to plan the pregnancy treatment process.