Performance Evaluation of Classifiers for Predicting Infection Cases of Dengue Virus Based on Clinical Diagnosis Criteria

Abstract

Dengue fever caused by dengue virus infection is a severe health threat that can lead to death. In the medical and health field, to classify data, data mining exploitation and classification methods have an essential role in predicting disease. Two main criteria are crucial to diagnosing dengue virus infection, namely the criteria clinical diagnosis and laboratory diagnosis. Dengue infection based on clinical signs and symptoms, as well as laboratory examinations, is made in three clinical diagnosis criteria, which consist of dengue fever (DF), dengue hemorrhagic fever (DHF), and dengue shock syndrome (DSS). This study was conducted with the primary objective to test and evaluate eight different classification algorithms to find the best algorithm in terms of efficiency and effectiveness. Classification algorithm used to predict dengue virus infection cases into three classes of DF, DHF, and DSS based on the performance of accuracy, precision, and recall. The classification algorithm used in this comparison were Neural Networks (NN), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree, Random Forest, Naïve Bayes, AdaBoost, and Logistic Regression. The dataset called DBDDKK was collected from the Division of Disease Prevention and Control in the Semarang City Health Office, Central Java, Indonesia. Impute missing values, selection relevant feature, and normalize feature conducted in the preprocessing stage resulted in 14,019 records with 16 attributes for each record. Then the data were split into 70% for training data and 30% for testing data. Cross-validation with the number of folds 10 is applied to validate the accuracy during the dataset training process. The result of the comparison shows that the NN algorithm has the best accuracy that was over other algorithms.