A successful treatment of outflow tract ventricular arrhythmia (OTVA) based on radio-frequency ablation requires a precise identification of the site of origin (SOO). Current methods in the clinics are based on qualitative analysis of pre-operative electrocardiograms (ECG), being heavily dependent on clinical expertise. Computational models have been proposed to assist OTVA procedures, but they are time-consuming and not adapted to clinical timings. We present here an alternative strategy to automatically predict the ventricular origin of OTVA patients using machine learning. We firstly trained machine learning models on simulated and patient ECG data using an XGBoost model. Feature relevance analysis showed that the most important feature for OTVA origin prediction was the amplitude of R-wave in V3, which coincides with the clinical criteria reported in literature. We then used the OTVA origin probabilities resulting from the XGBoost analysis to feed a Random Forest with patient clinical data (age, sex, hypertension) and ECG analysis (precordial transition and R-wave amplitude in V3), using automaticallydefined thresholds values, unless existing clinical cut-offs. Our proposed approach, tested on 85 patient cases, achieved an accuracy of 66% using the XGBoost model. When combining the probability per class with clinical data and automatic ECG analysis in the Random Forest model, the accuracy increased to 94%. This level of accuracy is comparable to existing clinical solutions, which have primarily been tested in single centers, relying on operator expertise and lacking generalizability. The developed automatic solution offers the advantage of real-time and generalized application. Additionally, the use of ML ensemble techniques, such as XGBoost and Random Forest, allowed for interpretation of the results, identifying the most relevant features for accurate origin prediction.