Analysis of Backpropagation Algorithm Using the Traingda Function for Export Prediction in East Java

Exports are significant for a country's economic development, especially in regions that carry out export activities because not all countries and regions have the same natural and human resources. Therefore this study aims to predict the level export of oil, gas, and others in the province of East Java and understand the forecast for the number of exports in the coming year. This is important to provide information to the East Java provincial government so that it can make policies so that the export value can be increased, at least so that the export value remains stable. The prediction algorithm used is the Backpropagation Neural Network algorithm using the Gradient Descent training function with Adaptive Learning rate. The research data is data on the Export of oil, gas, and others in East Java Province from 2008 to 2019. The prediction process analysis uses 3 network architecture models, namely: 5-10-1, 5-15-1, and 5-20-1. Based on the analysis results, the 5-10-1 model is the best compared to the other two models with an accuracy rate of more than 90% and MSE testing 0.0012454304, which means that this model is good for predicting the export of oil, gas, and others.