Thrombus formation in the left atrium (LA) is a major clinical complication associated with atrial fibrillation (AF) and diastolic dysfunction (DD). This study proposes a machine learning pipeline to predict thrombogenic regions using hemodynamic indicators derived from patient-specific Computational Fluid Dynamics (CFD) simulations. A dataset comprising eight simulations from distinct clinical scenarios (Normal, DDI, DDII, DDIII) and two anatomical models was used. Four features—ECAP, OSI, RRT, and TAWSS—were extracted and used to train and evaluate eight classifiers. XGBoost was selected as the best model based on Dice Score, using cross-validation and statistical analysis (Friedman and Wilcoxon tests). The final model achieved a Dice Score of $0.772 \pm 0.047$ on the test set. Evaluation across scenarios confirmed the model's robustness, and spatial visualizations enabled the identification of false positive and false negative regions. This approach enables high-throughput thrombus risk screening in patient-specific LA geometries and advances understanding of the hemodynamic correlates of thrombogenesis in DD and AF.