Background: Chronic Kidney Disease (CKD) is one of the growing non-communicable diseases in Iran, which, with the increasing prevalence of diabetes and hypertension, has significantly raised the demand for dialysis services. Managing limited resources in dialysis units (such as nurses, machines, and beds) under fluctuating demand and economic constraints is a major challenge for hospitals. Inaccurate demand forecasting and inefficient resource allocation lead to capacity waste, increased patient waiting times, and higher operational costs.
Aim: The aim of this study is to develop a hybrid machine learning and discrete-event simulation model to optimize hospital resource allocation in chronic kidney disease management. By predicting disease progression and future dialysis demand, the model enables dynamic performance evaluation and resource allocation optimization in dialysis units, ultimately enhancing operational efficiency and reducing costs.
Methodology: In the machine learning section, the Chronic Kidney Disease dataset from the UCI repository (containing 400 patients and 25 clinical features) was used, and four models—Random Forest, Support Vector Machine, Gradient Boosting, and CatBoost—were evaluated using Grid Search and TPE methods. In the simulation section, a discrete-event model of the dialysis unit at Shahid Ghalibaf Hospital in Bushehr was developed using Arena software, and various resource allocation scenarios (changes in nurse numbers, equipment, demand, failure rates, and patient arrival patterns) were tested. Findings: The Gradient Boosting model with Grid Search achieved the best performance with an F1-Score of 0.938, and the serum creatinine feature had the highest importance. In the simulation, the baseline scenario showed low nurse utilization (6.03%) and moderate machine/bed utilization (47.56%). The proposed scenarios produced limited changes, with no single scenario emerging as superior; for example, continuous arrival significantly increas