Prediction of University Patent Transfer Cycle Based on Random Survival Forest

Taking the invention patents of C9 League from 2002 to 2020 as samples, a random survival forest model is established predict dynamic time-point patent transfer cycle. By ranking variables based on importance, it found that countries citing, non-patent citations and backward have significant impacts C-index, Brier score integrated are used measure discrimination calibration ability four different models respectively. It prediction accuracy higher than Cox proportional risk model, lasso penalty model. In addition, function cumulative under adopted analyze individual university cycle, which shows has good performance able help universities well enterprises identify opportunities effectively, thereby shortening cycle improving efficiency.