Comparison study of neural network and deep neural network on repricing GAP prediction in Indonesian conventional public bank

The economic condition of a country is determined by the bank condition of that country it self, so that's why the whole world keep looking for the most effective and efficient method in conducting the assessment of the bank performance condition. Risk rate is commonly used to perform an analysis of the health condition of the bank, mainly using repricing gap. This research is propose deep neural network (DNN) which is one of deep learning algorithm as a method to predict the value of the repricing gap of the bank. The experiments is do a performance comparison of two methods that are neural network - standard backpropagation (SB) and DNN, by using 10 years historical data of monthly report. In DNN is using autoencoder which is an unsupervised algorithm as initiator value gradient on the formation of the network. The experimental results of both methods is the performance of DNN is better than SB. Using 30700 iteration, obtained the MSE value for DNN is more lower than SB. At high iteration amount, DNN still consistent to press the MSE value, and it's different with SB that start slowing down at the earliest iteration point, about 5700 iteration. DNN network topology that used in the experiments is the topology that can produce a model that have the most minimum MSE value, and that is equal network topology. (Every hidden layer neuron has the same number of units).