AI credit scoring system（japanese patent no. 7791512）

This technology uses a trained AI model based on decision-tree exploration through gradient boosting with LightGBM. It learns from credit data and rapidly determines the risk of delinquency from loan application data with high accuracy. It also delivers strong performance in fields where both high predictive accuracy, measured using metrics such as Accuracy and AUC, and accountability through factor explanation using SHAP are required.

Potential applications include: Medical diagnosis: Predicting the risk of disease onset and presenting the basis for the diagnosis using SHAP. Customer churn prediction: Detecting the risk of user cancellation in subscription services and analyzing the contributing factors.

Proof to date: The invention has been reduced to practice as a working implementation in Python (LightGBM + SHAP) and evaluated on loan application data. Measured performance: Accuracy ___ , AUC ___ . Every score is issued together with a SHAP-based attribution of the factors that produced it, so the reason for an individual decision can be presented to the applicant, to an internal audit function, or to a regulator. The patent is granted and registered in Japan (Patent No. 7791512). A packaged, runnable implementation is published as an Amazon Machine Image on AWS Marketplace, so a partner can evaluate the system in operation rather than on paper.

Next steps: We are seeking an exclusive licence (senyo-jisshiken), the strongest licence form under Japanese patent law: it is registered with the Japan Patent Office and is enforceable against third parties. Outright assignment of the patent is also available if the partner prefers full ownership. The patent right is effective in Japan; the technology itself is not territory-limited and can be deployed by the licensee in any market. The inventor is available for technical hand-over, model retraining on the partner's own data, and integration support.