US patent granted for machine-learning business performance prediction
US 12,591,902 B2 follows the 2023 Korean patent for the same method.
The United States Patent and Trademark Office has granted NFN patent US 12,591,902 B2, "Method for Predicting Business Performance Using Machine Learning and Apparatus Using the Same", with Lab Director Byung Jun Kang as inventor. It follows Korean patent No. 10-2573799, registered on 29 August 2023, and claims priority from Korean application 10-2022-0098623. The notice of allowance was reported in January 2026.
The method predicts company performance from structured data, such as financial statements and management indicators, together with unstructured data, such as reports, news and survey text. Text is cleaned, tokenized and encoded, turned into frequency-weighted feature vectors with a bag-of-words model, rescaled, and passed to an ensemble of classification and regression trees (CART).
US examination applies strict standards to software and business-method inventions. In allowing the claims, the examiner pointed to the combination of sparse-matrix data construction and the CART ensemble as a concrete technical application.
The method is the basis of DAL's business performance prediction research.
Original article: Money Today (Jan. 2026)