Comparative machine learning analysis identifies random forest and adaboost as superior models for the evaluation of semen quality and reproductive hormones.

Abstract

Background: Semen analysis is a widely accepted laboratory investigation for evaluating male infertility. However, in cases of idiopathic or unexplained male infertility, comprehensive assessment may require blood-based profiling of a reproductive hormone panel-including follicle-stimulating hormone (FSH), luteinizing hormone (LH), prolactin (PRL), and testosterone-to clearly identify endocrine abnormalities that may contribute to infertility. These variables exhibit nonlinear dynamics and multidimensional interdependencies, presenting analytical challenges for conventional statistical approaches to detect subtle, latent patterns within such complex data. In contrast, advanced analytics such as machine learning (ML) algorithms enable robust modelling and interpretation of high-dimensional datasets.

Methodology: Pre-processing of data included normalization and correlation analysis. Using a train-test split ratio of 80:20, nine ML algorithms-Linear Regression, Lasso Regression, Ridge Regression, Elastic Net, Random Forest, Support Vector Regression, Gradient Boosting, AdaBoost, and Neural Networks-were trained using cross-validation and hyperparameter optimization. Model performance was assessed using the coefficient of determination (R²), root mean square error (RMSE), and feature importance analysis.

Results: Ensemble learning approaches consistently outperformed conventional regression models. AdaBoost achieved the highest predictive accuracy for sperm motility (R² = 0.993), while Gradient Boosting yielded superior predictions for progressive motility (R² = 0.932) and vitality (R² = 0.982). Random Forest demonstrated the strongest performance for semen volume (R² = 0.428), LH (R² = 0.432), and PRL (R² = 0.586). Conversely, predictions for seminal pH (R² = 0.037), liquefaction time (R² = 0.158), FSH (R² = 0.178), and testosterone (R² = -0.114) were limited. Feature importance analysis identified total sperm count, sperm concentration, non-progressive motility, morphology, and non-motile sperm percentage as the most influential predictors across models.

Conclusions: Random Forest and AdaBoost emerged as the most effective and broadly applicable models for evaluating male reproductive parameters, whereas Gradient Boosting exhibited exceptional predictive capacity for select semen parameters.