Predicting patent lawsuits with machine learning

Abstract

We use machine learning methods to predict which patents end up in court using the population of US patents granted between 2002 and 2005. We show that patent characteristics have significant predictive power, particularly value indicators and patent-owner characteristics. Furthermore, we analyze the predictive performance concerning the number of observations used to train the model, which patent characteristics to use, and which predictive model to choose. We find that extending the set of patent characteristics has the biggest positive impact on predictive performance. The model choice matters as well. More sophisticated machine learning methods provide additional value relative to a simple logistic regression. This result highlights the existence of non-linearities among and interactions across the predictors. Our results provide practical advice to anyone building patent litigation models, e.g., for litigation insurance or patent management more generally.