Performance Comparison of Machine Learning Models for the Prediction of Dialysis Treatment Variables

Abstract

The increasing incidence of kidney disease has led to a growing number of patients requiring dialysis, a treatment essential for blood purification. However, dialysis is often associated with complications that are difficult to anticipate using traditional models, a challenge for which Artificial Intelligence offers promising support. This study represents a first step towards developing a machine learning model capable of predicting optimal dialytic operational parameters, which clinicians can use to achieve the desired clinical outcomes for each patient. Using data from ~ 250 dialysis sessions, fifteen models were tested and compared. Among these, the Gradient Boosting and the eXtreme Gradient Boosting algorithms exhibited the highest predictive accuracy for blood flowrate and heparin volume (R2 > 0.65), showing the potential of data-driven models to enhance patient safety and dialysis care.