This study presents a machine-learning approach for diagnosing vehicle automatic transmission failures using simulation data from a KIA-MORNING Si AT 2016 model. The dataset, generated through a comprehensive Simulation-X model of the powertrain system, includes three key vehicle dynamics features: acceleration time from 0 to 100 km/h (t100), vehicle speed after 30 s (v30), and power output after 30 s (P30). Twelve categories of transmission failures, including mechanical and hydraulic faults of clutches and brakes, torque converter failures, and standard operational conditions (low power loss or hydraulic leakage), were simulated. The study reviewed and employed Support Vector Machines (SVM) for classification, with hyperparameter tuning of SVM (C and γ), achieving 92,5% test accuracy. The SVM model demonstrated perfect classification in seven failure classes and high accuracy in most others, though some overlap caused misclassification between similar failure types and normal operations. Feature importance analysis highlighted the dominant role of acceleration time from 0 to 100 km/h, followed by v30 and P30. The results indicate that the proposed SVM-based diagnostic model can effectively classify automatic transmission failure types, supporting maintenance decisions. Future work includes exploring other machine learning algorithms to improve classification accuracy and regression modelling further to quantify the power loss and hydraulic leakage.