Roughly thirty percent of coronary artery bypass graft (CABG) patients develop atrial fibrillation (AF) in the five days following surgery, increasing the risk of stroke, prolonging hospital stay three to four days, and increasing the overall cost of the procedure. Current pharmacologic and nonpharmacologic means of AF prevention are suboptimal, and their side effects, expense, and inconvenience limit their widespread application. An accurate method for identifying patients at high risk for postoperative AF would allow these methods to be focused on the patients on which its utility would be highest. The main objective of this research was to develop a Bayesian network (BN) which could model/predict/assign risk of the occurrence of atrial fibrillation in CABG patients using retrospective data. A secondary objective was to develop an integrated framework for more advanced methods of feature selection and fusion for medical classification/prediction. We determined that the naïve Bayesian network classifier used with features selected by a genetic algorithm is a better classifier to use, given our cohort. The naïve BN allows for reasonable prediction despite being presented with patients with missing data points as might occur in the hospital. This classifier achieves a sensitivity of 0.63 and a specificity of 0.73 with an AUC of 0.74. Furthermore, this system is based on probabilities that are well understood and easily incorporated into a clinical environment. These probabilities can be altered based on the cardiologists prior knowledge through Bayesian statistics, allowing for online sensitivity analysis by doctors, to perceive the best treatment options. Contributions of this research include: - An accurate, physician-friendly, postoperative AF risk stratification system that performs even under missing data conditions, while outperforming the state of the art system, - A thorough analysis of previously examined and novel pre- and postoperative clinical and ECG features for postoperative AF risk stratification, - A new methodology for genetic algorithm-built traditional Bayesian network classifiers allowing dynamic structure through novel chromosome, operator, and fitness definitions, and - An integrated methodology for inclusion of doctor s expert knowledge into a probabilistic diagnosis support system.