Atherosclerosis (AS) is a chronic inflammatory vascular disorder in which endoplasmic reticulum stress (ERS) plays a crucial regulatory role. However, the biological and translational relevance of ERS-related gene networks in AS remains largely unexplored. This study aimed to identify a robust ERS-related gene signature for AS. We integrated multiple GEO datasets and applied machine learning algorithms, including least absolute shrinkage and selection operator (LASSO) regression, support vector machine-recursive feature elimination (SVM-RFE), and random forest (RF). Five ERS-related signature genes (TRIM25, CYBB, CYBA, MYOC, and PRKAA2) were identified and showed favorable discriminatory performance in the integrated discovery cohort (combined AUC = 0.946). The expression patterns of these genes were further examined at both the mRNA and protein levels by quantitative real-time polymerase chain reaction (qRT-PCR) and Western blotting (WB) in an oxidized low-density lipoprotein (ox-LDL)-induced endothelial injury model. Gene set enrichment analysis and immune infiltration analysis indicated that the identified genes were primarily involved in oxidative stress and immune-related pathways. Collectively, this study identifies a machine learning-derived ERS gene signature associated with AS. These findings improve our understanding of ERS-related vascular injury in AS and provide candidate biomarkers for further tissue-level and mechanistic validation.