Background: Breast cancer is the most common cancer in women, and the lung is one of the frequent sites for metastasis in breast cancer. Identifying biomarkers to predict lung metastasis is essential for early detection and targeted intervention, thereby improving survival rates for patients with lung metastatic breast cancer. Method: Four datasets (E-MTAB-365, GSE2603, GSE11078, and GSE14020) from NCBI and Array Express databases, containing gene expression profiles and clinical data from breast cancer patients, were selected. High-throughput screening identified potential biological markers by evaluating the predictive ability of each gene for lung metastasis, and a Venn diagram was used to find common genes across datasets with an AUC > 0.65. Using the linear regression algorithm, we developed a gene-based model to predict the risk of lung metastasis with E-MTAB-365 as the training dataset and the remaining datasets for validation. Model performance was evaluated through ROC curve analysis and the Kaplan-Meier curve. Results: The model, consisting of three genes (IRAK1, ATP11A, and LYN), achieved a good AUC across the analyzed datasets. The model demonstrated that patients with lung metastasis had significantly higher risk scores than those without. In addition, patients with high-risk scores faced a higher risk of lung metastasis and a shorter non-metastatic survival time. Conclusion: We successfully built a machine-learning model based on gene expression to predict lung metastasis in breast cancer patients and validated its reliability. Our results suggest the potential application of the established model as a predictive tool to assist physicians in practical diagnosis.