• Identified 16 autophagy-related genes as potential biomarkers for recurrent spontaneous abortion (RSA) through LASSO and logistic regression analysis.
• Constructed a nomogram model with good predictive accuracy, validated by ROC curves and decision curve analysis.
• Selected an optimal neural network model among six machine learning models, highlighting MAP2K7, CALCOCO2, SAR1A, TUSC1, and STK11 as key variables.
• Enrichment analyses revealed autophagy-related genes are involved in autophagy regulation, PI3K/Akt signaling, and metabolic processes, providing insights into RSA pathogenesis.