Application of machine learning using hormonal biomarkers in male infertility prediction: a systematic mapping review

male factors contribute to approximately 40–50% of infertility cases, yet diagnostic methods remain invasive, subjective, and often inconclusive. Hormonal biomarkers such as FSH, LH, and testosterone are routinely assessed but have limited predictive value when used alone. Recent advances in machine learning (ML) have demonstrated potential to enhance diagnostic precision through pattern recognition in complex hormonal data. This systematic mapping review aimed to identify, characterize, and synthesize studies applying machine learning algorithms using hormonal biomarkers for predicting male infertility and related outcomes. Following the PRISMA–ScR guidelines, a systematic search was conducted in PubMed, Scopus, and Web of Science databases up to May 2025. Eligible studies included peer-reviewed articles applying ML models based on serum hormones to predict male infertility or related outcomes, and reporting quantitative performance metrics. Studies not involving ML methods, hormonal predictors, or male-specific analyses were excluded. Data were extracted on study design, hormonal features, algorithms used, and model performance, and synthesized narratively due to heterogeneity across datasets and methods. Out of 115 retrieved records, 11 studies met the eligibility criteria. Most studies were retrospective and hospital-based, with sample sizes ranging from under 150 to over 1,000 participants. Geographically, Turkey and China dominated research contributions, focusing primarily on sperm retrieval success, infertility risk assessment, and OA–NOA differentiation. The most frequently used ML algorithms were Artificial Neural Networks (ANN; n = 6), Support Vector Machines (SVM; n = 4), k-nearest neighbors (kNN; n = 4), and ensemble methods, such as XGBoost, CatBoost, and super learner, which demonstrated superior predictive accuracy in several datasets. Reported accuracies typically exceeded 80%, with AUC values between 0.80 and 0.90, though sensitivity and specificity varied depending on outcome measures. Core hormonal predictors—FSH, LH, and total testosterone (TT)—were included in nearly all studies due to their physiological relevance in spermatogenesis. Emerging biomarkers such as Anti-Müllerian Hormone (AMH), Inhibin B, and leptin improved model performance and biological interpretability in select analyses. Studies that combined multiple hormones achieved stronger predictive capacity than single-marker models. This systematic mapping review demonstrates that machine learning models leveraging hormonal biomarkers—particularly multihormonal combinations, including FSH, LH, testosterone, and emerging markers, such as AMH and inhibin B—show strong potential for non-invasive prediction of male infertility. Artificial neural networks and ensemble-based approaches consistently achieved robust predictive performance across diverse clinical contexts. Despite these promising findings, the current evidence base is largely retrospective and geographically concentrated. Future research should prioritize prospective, multicenter validation and integration of explainable AI techniques to enhance clinical transparency, generalizability, and real-world applicability. PROSPERO Registration ID: CRD420251229798.