The mortality rate is one of the important aspects in determining insurance premiums. The mortality rates have influenced by several factors, i.e., air quality. Therefore, we consider Deep Neural Network (DNN) model for prediction of the air quality-based mortality rate. In this paper, we examine two DNN architectures. The first architecture consists of five layers including an input layer, a hidden layer, two hidden dropout layers, and an output layer. The second architecture consists of four layers including an input layer, a hidden layer, a hidden dropout layer, and an output layer. We optimize dropout rates and activation functions to obtain the optimal accuracies. Our simulations show that the first DNN architecture produces a slightly better performance. The DNN architecture uses ReLu as activation function and applies a 40% dropout rate for both dropout hidden layers. This DNN architecture also gives slightly better accuracy than the standard one hidden layer Neural Networks.
|Journal||Journal of Physics: Conference Series|
|Publication status||Published - 17 May 2019|
|Event||2nd International Conference on Data and Information Science, ICoDIS 2018 - Bandung, Indonesia|
Duration: 15 Nov 2018 → 16 Nov 2018