Patent lifetime prediction using LightGBM with a customized loss.

Area of Science:

• Intellectual Property Management
• Machine Learning Applications
• Data Science

Background:

• Patent lifespan is a key metric for assessing economic value and business potential.
• Accurate patent duration forecasting is crucial for patent holders and strategic decision-making.
• Existing research often focuses on patent stages, not individual patent commercial viability through lifespan.

Purpose of the Study:

• To develop a precise method for predicting the probability of a patent remaining valid until its maximum expiration date.
• To assess the commercial viability of individual patents by analyzing their predicted lifespan.
• To improve upon existing patent assessment methodologies by focusing on duration prediction.

Main Methods:

• Utilized a dataset of 200,000 patents for evaluation.
• Implemented a machine learning approach combining LightGBM with a customized Focal Loss function.
• Developed a novel loss function derived from Focal Loss to enhance prediction accuracy.

Main Results:

• The combined LightGBM and Focal Loss model demonstrated significant performance improvements.
• Incorporating Focal Loss enhanced the model's ability to prioritize difficult-to-classify instances during training.
• The targeted approach improved the model's accuracy in distinguishing samples and recovering from challenges.

Conclusions:

• The proposed method effectively predicts patent lifespan, offering a valuable tool for commercial viability assessment.
• Combining LightGBM with Focal Loss significantly enhances prediction accuracy for patent duration.
• This approach provides a more accurate and robust method for patent value analysis.