We use machine learning methods to predict which patents end up at court using the population of US patents granted between 2002 and 2005. We analyze the role of the different dimensions of an empirical analysis for the performance of the prediction - the number of observations, the number of...