A novel non-invasive method for prognosis of Dengue fever

Prediction of Dengue presents great challenge as the clinical symptoms overlaps with other conventional fever. Dengue viral infection has been reported in more than 100 countries, with total of 2.5 billion people. Testing of prognosis, the stages of dengue such as DF, Dengue Hemorrhagic Fever (DHF) and Dengue Shock Syndrome (DSS) under conventional methods needs frequent blood test information. The condition of dengue infected patient becomes more ill from one stage to another and continuous blood test will trouble the patients in terms of wastage of blood and make them restless. In the proposed noninvasive method, the Generalized Regression Neural Network (GRNN) was trained with existing patient's blood test database in order to perform diagnosis and prognosis of stages. While analyzing the database, for training process it was observed that there is occurrence of missing data in the database that widely affects the training process. The database is classified using adaptive decision tree method to improve the imputation accuracy. In order to fill the missing data, K-Nearest Neighbors (K-NN) imputation method is adapted. The performance of the trained GRNN for early diagnosis of DF and prognosis of stages from the instant blood test information of test patient had been analyzed with actual results and trained Radial Basis Function Neural Network (RBFNN), for adult and child database. It is observed that K-NN provides 95% accuracy; GRNN provides 95% accuracy and 0.875 Area Under the Curve (AUC) for prognosis and diagnosis.