EN
When patients return to the emergency department (ED) within 72 hours after their previous ED discharge, it is generally assumed that their initial evaluation or treatment had been somehow inadequate. Mining data related to unplanned ED revisits is one method to determine whether this problem can be overcome, and to generate useful guidelines in this regard. In this study, we use the receiver operating characteristic (ROC) curve to compare the data mining model by affinity set to other well known approaches. Some scholars have validated the affinity model for its simplicity and power in handling information systems especially when showing binary consequences. In experimental results, SVM showed the best performance, with the affinity model following only slightly behind. This study demonstrated that when patients visit the ED with normotensive status or smooth breath patterns, or when the physician-patient ratio is moderate, the frequency with which patients revisit the ED is significantly higher.