Forecasting Iran Saffron Export by Comparison of Machine Learning Algorithms

Imports and exports in all countries play an important role in economic growth. Therefore, choosing the right products will increase the country's competitiveness in world trade. Saffron is one of the most important and distinctive non-oil products of Iran for export. The purpose of this study is to predict the export of saffron through three machine learning algorithms and select a proper algorithm for predicting. The sample period of the forecasting models includes the data of Iran Saffron export from 2012 to 2019 which have been collected from the Iran Saffron Association. After performing the data preparation steps, the saffron export prediction was performed using three data mining algorithms including artificial neural network, deep learning, and gradient boost tree. To choose a better forecasting model, the validity of the model plays an essential role. Predictive validity of three designed models, using Absolute Error (Artificial neural network = 0.036, Network deep learning = 0.031, Gradiant boost tree = 0.047), R-squared (Artificial neural network = 0.045, Network deep learning = 0.044, Gradiant boost tree = 0.073) and Correlation (Artificial neural network = 0.95, Network deep learning = 0.98, Gradiant boost tree = 0.97) were measured. Based on the findings, all of these three designed models are accurate and their prediction error is very low and close to each other. However, with insignificant differences, the deep learning network has less error. The results could be useful for more accurate planning of saffron exports.