The environmental costs of intensive farming activities are often under-estimated or not included into rural development plans, even though they play an important role in addressing future society's needs. This paper focuses on the use of statistical learning methods to predict N2O emissions and N leaching under several conservative scenarios, in order to provide an alternative approach to deterministic models on a macro-scale. To that aim, three learning methods, namely neural networks (multilayer perceptrons), SVM and random forests, are compared and provide accurate solutions.