An Explainable Machine Learning-Based Approach to Patent Litigation Risk Assessment

Machine learning-based patent litigation prediction has gained attention as a means of assessing patent litigation risk proactively. However, prior studies have faced limitations due to insufficient model interpretability and inadequate incorporation of the multifaceted technological and organizational contexts underlying patent litigation. This study proposes an explainable machine learning-based framework for patent litigation risk assessment that integrates technological attributes and applicant characteristics while ensuring interpretability. The framework comprises four stages: (1) patent and litigation data collection from the USPTO(United States Patent and Trademark Office) and PACER Public Access to Court Electronic Records), (2) extraction of 16 technology-level and 5 organization-level indicators, (3) a two-stage prediction model—litigation occurrence classification followed by severity assessment—with comparative evaluation of machine learning algorithms including Random Forest, XGBoost, and LightGBM, and (4) stage-specific key determinant identification through SHAP(Shapley additive explanations) analysis. Empirical analysis using 33,603 patents and 81,335 litigation records demonstrates effective prediction of technology infringement risk, with key determinants consistent with existing literature. A risk classification scheme is further introduced to support strategic decision-making in intellectual property management.