A novel approach to evaluating the business potential of intellectual properties: A machine learning-based predictive analysis of patent lifetime

Abstract

Because most patent holders are likely to retain their patents only when the profits created are larger than the maintenance fee, the patent lifetime has been used as a proxy of a patent’s business potential in a patent evaluation. In this paper, we propose an approach to evaluate the business potential of individual patents by applying a machine learning algorithm to predict the likelihood that a patent will survive until its maximum expiration date. A total of 24 indicators and historical maintenance fee event data obtained from a large number of patents were used for learning a feed-forward neural network (FFNN) model. Consequently, the FFNN model outperformed other algorithms and interesting trends were identified, namely, the recall is higher than precision irrespective of the models used, and the predicted business potential grade of a patent has a strong positive correlation with the actual lifetime of the patent. An advantage of the approach is its wide applicability in dealing with patents with few or no forward citations. This approach can be employed for examining patents with high business potential to uncover hidden core technologies of competitive firms or identifying newly issued patents with a high business potential to predict emerging technologies.