Comparison of activation functions on radial basis function neural network in predicting dengue hemorrhagic fever incidents in DKI Jakarta

Abstract

Dengue Hemorrhagic Fever (DHF) is still one of the main public health problems in Indonesia. The number of DHF cases and the spread of this disease is increasing along with mobility and population density. The Radial Basis Function Neural Network (RBFNN) in this research was implemented to predict the number of weekly DHF incidents in DKI Jakarta. RBFNN is a Feed Forward Neural Network model that has a single hidden layer. The hidden layer of RBFNN is constructed by an activation function. K-means clustering algorithm is used to improve the performance of RBFNN to determine the center and width of the activation function. This research used two different activation functions, which are the Gaussian and multiquadratic functions. The final results of this research indicate that the selection of the best activation function depends on the data and RBFNN structure used. The value of MSE contributes to the graph of the predicted results of weekly DHF incidents.