Prediction of Length of Stay and Cost for Diabetes Inpatient using C4.5 Algorithm

Abstract

Currently, Electronic Medical Record (EMR) is not very informative for extracting useful information or tracking a patient’s illness. Using data mining techniques, such confidential information can be extracted by finding patterns and correlations between attributes that help practitioners understand hidden patterns. One technique that is widely used in data mining is a decision tree. Decision trees are useful for exploring data finding hidden relationships between several candidate input variables and a target variable. Many algorithms can form decision trees, including the C4.5 Algorithm. The process in the decision tree is changing the shape of the data table into a tree model, changing the tree model into rules, and changing the rules. This can be used to make assumptions regarding class names, classifying knowledge about training sets and class labels, and classify newly acquired data. Although many studies have discussed predicting diabetes, they only focus on two label, namely predicting diabetes or not. This study will predict the range of days of stay and the cost of treating diabetes patients using the Decision tree method with the C4.5 Algorithm. After implementation, the next step is to calculate the accuracy-test with different training and test data. The test is carried out using the Confusion Matrix with an accuracy of 85,84%. This study also suggests adding the variables of diabetes symptoms, disease history, and other administrative costs that will be a supporting factor to determine the length of stay and the cost of treatment for diabetes patients.