Modeling backpropagation neural network for rainfall prediction in tengger east Java

Climate Change in the world gives many impacts on changing rainfall patterns. As a result, potato farming areas such as Tengger, East Java are having problems to determining potato growing times. Because of that, the production of potatoes decreased. It needs a method that can predict rainfall based on rainfall patterns that occur after climate change. Backpropagation Neural Network (BPNN) is one method that can learn from the data of the past and make it a guide to predict future data. In this research, BPNN is used to predict rainfall in Tengger, East Java using rainfall data from 2005 to 2014 which is the period of climate change. Data from 2005–2009 is used as training data and data from 2010–2014 use as testing data. In addition, this research is also looking for the most optimum parameter modeling of BPNN includes the value of learning rate, hidden layer, and maximum epoch. Based on the test results, the most optimum parameter of BPNN that is using learning rate 0.4, hidden layer 3, and maximum epoch 4000. From the optimum model of BPNN, the result of average error RMSE for 4 location in Tengger is 8.14 from training process and 8.28 from the testing process. The result of RMSE error with BPNN is smaller compared to previous research using Tsukamoto Fuzzy Inference System (FIS).