<h3>Purpose/Objective(s)</h3> Patients with locally advanced non-small cell lung cancer (NSCLC) are at increased risk of developing major adverse cardiac events (MACE) following radiotherapy. Our group previously identified cardiac substructure dose constraints associated with increased risk of MACE. We employ a machine learning (ML) approach to further identify predictors of MACE and evaluate ML capacity for personalized MACE risk-stratification. <h3>Materials/Methods</h3> Retrospective analysis of 701 patients with locally advanced NSCLC treated with thoracic radiotherapy between 2003-2014. MACE included unstable angina, heart failure, myocardial infarction, coronary revascularization, and cardiac death. We used extreme gradient boosting for MACE prediction. Input features included 195 demographic and 240 radiation dose and/or anatomic. Cardiac substructures were manually delineated; heart and chamber volumes were indexed to body surface area (BSA). We split data 70/30 into training/testing, balanced by MACE. Hyperparameters were bootstrap-tuned with 50-round grid search. Area under the receiver operator characteristic curve (AUC) evaluated model performance. We trained our model on any MACE regardless of timeframe and tested performance at varying timepoints (90-day, 180-day, etc.). Shapley values measured feature importance and were used to construct personalized risk profiles for each patient. <h3>Results</h3> Among 701 patients, 70 developed ≥1 MACE. The median age was 65 and median time to first MACE was 20.6 months. Training AUC for any MACE: 0.67; testing AUC: 0.73. Our model displayed time dependent performance improvement, with high accuracy for MACE closer to radiation start. Testing AUC for 90-day MACE: 0.97; 180-day MACE: 0.94. Table 1 displays additional AUCs. The top 10 predictive features for MACE were coronary heart disease history, right atrial volume, left circumflex coronary artery (CA) volume (V) receiving 15 Gy (V15Gy), heart volume, lung V55Gy and V70Gy, hypertension, left ventricle V15Gy, left main CA volume, and left anterior descending CA V15Gy. Based on patients' Shapley-derived risk profile, radiation dose variables generally became less predictive of MACE as time from radiation increased, while demographics remained generally predictive. <h3>Conclusion</h3> ML modeling enabled high precision for predicting short-term MACE in locally advanced NSCLC patients who received radiotherapy, though long-term MACE was less accurate. ML techniques show promise for identifying patients at high risk of short-term MACE. Our Shapley-derived individual risk profiles may assist with baseline cardiac risk assessment and identify opportunities for risk mitigation. Prospective validation may help implement these tools into clinical practice.