Intelligent diagnosis‐related group classification for 134,189 patients in eastern China using machine learning

Abstract

Diagnosis‐related group (DRG) classification is crucial for healthcare cost management and resource allocation, but traditional manual classification by physicians is inefficient and error‐prone, especially for large‐scale medical data. To address this challenge, we propose ELGWO‐LightGBM (elite gray wolf optimization–LightGBM), an automated DRG classification method that integrates an enhanced gray wolf optimization (GWO) algorithm with LightGBM. The method incorporates an elite wolf mechanism (ELGWO) to optimize LightGBM hyperparameters, effectively avoiding local optima while improving convergence speed. We evaluated our approach on a dataset of 134,189 patient records across 57 DRG categories. Experimental results demonstrate that ELGWO–LightGBM significantly outperforms traditional deep learning models and gradient boosting methods in classification accuracy. The ELGWO optimization also enhances model stability and robustness when handling large‐scale structured medical data. Our results show that ELGWO–LightGBM provides an efficient and accurate automated solution for DRG classification, contributing to the advancement of intelligent healthcare data processing systems.