Comparing activation functions in predicting dengue hemorrhagic fever cases in DKI Jakarta using recurrent neural networks

Abstract

Dengue hemorrhagic fever (DHF) is a disease caused by the dengue virus and spread by infected Aedes aegypti and A. albopictus mosquitoes. Various socio-economic and environmental factors make it difficult to predict DHF incidents. However, with machine learning, we can make more accurate predictions based on historic data. The spread of DHF in a given region can be predicted based on incident data. In this research, one means of machine learning, the Recurrent Neural Network (RNN), is used to predict DHF incidents in DKI Jakarta by using historic DHF case data from 2009 to 2017. RNN is a neural network with a recurrent hidden state which is activated using current data and previous data. RNNs are well-suited to predicting time-series data. In the implementation, we use three activation functions that is sigmoid, tanh, and ReLU to determine which one is the most accurate in predicting DHF incidents in Jakarta. The implementation results show that the sigmoid activation function can give better results on the RNN model compared to tanh and ReLU activation functions.