一件專利從申請、獲證並持續維護，需花費很高的成本，獲證後的專利權不僅是發明人心力成果，也是大學珍貴的無形資產。然而專利獲證後若無授權，長期繳納維護費用，著實為大學的負擔，如何在資源有限下，優先維護價值較高較具商業淺力的專利，並提高管理決策的時效，為大學裡長遠課題，也是本研究的初心。以國立中央大學為研究個案，彙整學校專利資料並利用迴歸分析，篩選出影響專利價值的因素，將這些因素使用不同的預測模型，從中選擇表現較佳且穩定的梯度提升模型。透過模型將專利分成四個等級，並依分級結果，建議專利維護優先順序，並選出具商業淺力之專利進行推廣運用，使有限的經費獲得運用並發揮效益。實證結果顯示，透過機器學習模型可有效的將專利評分轉為價值評等，縮短判斷時效，期盼學術機構得以應用此模型降低維護成本，提升研發成果的實用性。;Maintaining a patent from application to grant and throughout its lifecycle incurs substantial costs. Once granted, a patent not only represents the inventor’s intellectual effort but also serves as a valuable intangible asset for universities. However, patents that remain unlicensed after being granted can become a financial burden due to recurring maintenance fees. Therefore, how to prioritize the maintenance of high-value patents with greater commercial potential under limited resources, while improving the timeliness of management decisions, has become a long-term challenge for universities and the driving motivation of this study.This research uses National Central University as a case study. By compiling the university’s patent data and applying regression analysis, we identify key factors influencing patent value. These factors are then incorporated into various predictive models, from which l Gradient Boosting is selected for its superior and stable performance. The model categorizes patents into four levels and, based on these classifications, provides recommendations for maintenance priority and highlights patents with commercialization potential for promotion and utilization. This approach aims to maximize the use of limited funding and enhance the overall impact of university-owned patents.Empirical results have shown that machine learning models can effectively evaluate patent value, significantly reducing decision-making time. It is hoped that academic institutions can adopt this model to lower maintenance costs and improve the practical application of research outcomes.
