Context: Coronary artery disease (CAD) remains a leading cause of mortality worldwide. Coronary CT angiography (CCTA) enables non-invasive plaque assessment, while pericoronary adipose tissue (PCAT) reflects vascular inflammation. Radiomics, combined with machine learning (ML), has shown promise in identifying imaging biomarkers of plaque vulnerability and predicting MACE; however, the integrated radiomic characterization of both plaque and PCAT remains largely unexplored. This study aimed to evaluate whether CCTA-based radiomic features derived from coronary plaques including PCAT can stratify patients according to MACE risk using ML modelling. Methods: A retrospective cohort of 194 patients, comprising 685 plaques, was analyzed. Regions of interest including PCAT were defined, and 572 radiomic features were extracted using PyRadiomics. Feature robustness was assessed through morphological perturbations, retaining only stable features (ICC > 0.75). A ML pipeline incorporating feature selection (Mann–Whitney U test, Spearman correlation filtering, and LASSO) and classifier optimization was implemented to identify high-risk MACE patients. Seven classifiers were evaluated, and patient-level predictions were obtained by aggregating plaque-level outputs using a worst-case (maximum probability) strategy. Model performance was assessed on a hold-out test set (39 patients, with confidence intervals estimated via bootstrap resampling). Results: The best-performing selector–classifier combination, defined as the one achieving the highest mean balanced accuracy during a 3-fold cross-validation, was a multilayer perceptron with a correlation threshold of 0.85. This model achieved a balanced accuracy of 0.848 (95% CI: 0.815–0.879) and an AUC of 0.833 (95% CI: 0.768–0.894) on the hold-out test set. Notably, the model achieved 100% sensitivity, correctly identifying all MACE patients, with a specificity of 0.697. Conclusion: These findings demonstrate that radiomic features derived from plaques including PCAT enable effective stratification of patients at risk of MACE. This approach shows promise for improving personalized risk assessment in CAD, although validation in larger, multicenter cohorts is required.