Up Sampling Data in Bagging Tree Classification and Regression Decision Tree Method for Dengue Shock Syndrome Detection

Dengue virus, DENV, is the cause of dengue fever and carries the risk of developing dengue hemorrhagic fever and dengue shock syndrome (DSS), which can lead to death. The high mortality rate is due to the blockage of blood circulation resulting from delayed patient management, particularly in cases of advanced-stage DSS. Intensive examinations and new treatments are only provided when the patient has been admitted to the hospital. Therefore, we proposed to develop an early detection for DSS risk based on the clinical data using ensemble method through bagging Tree-CART, namely Tree-Bag algorithm. We explored information from patient’s clinical data including temperature, vomiting, pain, as well as the levels of erythrocytes, leukocytes, creatine level, hemoglobin, and time series data on the patient's condition. In the preprocessing data stage, we applied oversampling data due to imbalanced class in dataset. The experimental results show that the Tree-Bag achieved high performance with accuracy of 0.91 and the AUC of 0.84. It is dominant when compared to single machine algorithms including: KNN, Naïve Bayes, CART decision tree, and SVM.