Patent lifespan prediction and interpreting the key determinants: An application of interpretable machine learning survival analysis approach

While the lifespan of patents is widely regarded as a key indicator for assessing their economic value, its utility in patent valuation is significantly constrained, as it can only be accurately measured at the time of patent expiration. Addressing this limitation necessitates proactively predicting the expected patent lifespan and thoroughly analyzing the complex relationships among various factors that affect patent lifespan. In response, this study constructs an interpretable machine learning framework to predict patent lifespan and explores the factors influencing it. The framework integrates features from five dimensions: technical, legal, market, patentee, and textual. It develops five distinct machine learning survival analysis models and employs post-hoc interpretable machine learning techniques on the optimal model to investigate the intricate relationships between these features and patent lifespan. The results of an empirical study of patents in China's Yangtze River Delta region demonstrate that the machine learning survival analysis approach significantly outperforms the traditional Cox proportional hazards model (Cox-PH) in terms of predictive performance. Furthermore, the post-hoc interpretation technique provides precise descriptions of the effects of various features on patent lifespan, revealing previously unidentified nonlinear relationships. This study holds substantial significance for the research and application of patent valuation, early patent warning, patent pledge financing, and patent management.