比較多種機器學習預測技術於急診看診人數預測之研究

在醫院急診室內，需要進行急迫性醫療的患者，都在等待接受緊急治療，所以急診室的時間每分每秒都很寶貴。由於健保制度造成各醫院急診擁塞情形相當嚴重，緊急醫療資源有限，如果我們能預測急診室會出現多少患者，那麼醫院就能更有效調度資源，也能減緩醫護人員在急診時的壓力，提升醫療品質，急診病患的等待時間也就會減少，所以急診的醫護人員要如何有效的分配資源，提供給患者最優質的服務是一個值得探討的議題。本研究使用方法為Weka系統裡的Simple Linear Regression、Linear Regression、Multilayer Perceptron、SVR、RBF Network、IBK、K Star、LWL、REP Tree、M5 Rules等十種預測模式，目的在於比較多種機器學習法的預測技術，透過機器學習法找出合適於預測急診人數的預測技術，希望作為急診的人力資源與醫療資源分配的依據。研究結果顯示SVR 與REP Tree預測模式是較適合提供給地區型醫院用來預測急診人數的預測方法。

關鍵字

急診預測；機器學習；支援向量迴歸；快速決策樹

並列摘要

In the emergency room of the hospital, patients who need emergency medical care are waiting for emergency treatment, so the emergency room time every minute is very valuable. Because the health insurance system is very serious and the emergency medical resources are limited, if we can predict how many patients will appear in the emergency room, the hospital will be able to dispatch the resources more effectively, but also reduce the pressure of the medical staff in the emergency, improve the medical Quality, emergency patients waiting time will be reduced, so the emergency medical staff how to effectively allocate resources to provide patients with the best quality service is a question worthy of discussion. This study uses ten prediction modes such as Simple Linear Regression, Linear Regression, Multilayer Perceptron, SVR, RBF Network, IBK, K Star, LWL, REP Tree and M5 Rules in Weka system. The purpose of this paper is to compare the prediction techniques of machine learning Method to find out the forecasting technology suitable for predicting the number of emergencies, hoping to be the basis for the allocation of human resources and medical resources. The experimental results show that the SVR and REP Tree prediction models are more suitable for the prediction of the number of emergencies provided by the regional hospitals.