Neural networks with Turkey's export value forecast

The concept of foreign trade, which has a history of about two and a half centuries and which is of great importance for the open economies, plays a key role in the development and development of national economies in today's global competition environment. Exports have an important place in foreign trade transactions. Because, there is a driving force between export, economic growth, foreign exchange, and capital inflow. This study, the first in Turkey with the aim to identify the variables that affect the export using annual data covering 2002-2017 dates for the structural equation model was established. According to the studies in the literature, independent variables such as gross capital formation, industry, savings, exchange rate, logistics, GDP, per capita GDP, commercial service, and export of goods are used. As a result of the analysis, it was determined that the independent variables explained the export dependent variable as high as 99%. It was observed that the model's CR, AVE and Cronbach's Alpha values exceeded the threshold values, thus satisfying the reliability, and internal validity of the model. While the independent variables of commercial service and goods export have a positive and meaningful effect on the export dependent variable, it is seen that the logistic variable has higher priority for the export dependent variable compared to other independent variables. The research model was found to have a high generalizability of 99%. In the last part of the analysis, a model was established with artificial neural networks using the same variables. Estimates were calculated with 2 - 9 hidden layers in the model established with artificial neural networks and the real values and predicted values were compared by calculating the values of R2, RMSE, MAE and MAPE (%). In the model, in the 9 hidden layers R2 = 0.99, RMSE = 3611616, MAE = 2613313 and MAPE = 45,14141%. this shows that the best result is the number of hidden layers 9. As the number of hidden layers of the network increased, statistically significant results were obtained. Keywords: Export, Factors Affecting Exports, Structural Equation Model, Artificial Intelligence