Predicting the import and export of commodities using Support Vector Regression and Long Short-Term prediction models

Abstract

The prediction of import and export of commodities have occurred between countries to either buy or sell goods essential for humans. Governments need to keep track of the amount of import or export to ensure the increase of Gross Domestic Product (GDP) for their country. Support Vector Machine (SVM) is a powerful classification algorithm to classify data efficiently. Support Vector Regression (SVR) is a modification of SVM that predicts absolute values. The purpose of this paper is to use SVR in a commodity dataset to predict each commodity's price being imported and exported for limited countries. SVR uses the support vectors obtained during the running of the algorithm to predict the dataset's outcome. The new version of SVR algorithm is proposed which is assisted with modified RBF Kernel to improve the model's efficiency. Further LSTM is applied for prediction in layers to predict the weight of some incoming commodities to countries. We then obtain the predicted results and find the accuracy of the model using this result over a real dataset. The results show that the over-all error for the proposed model is very trivial and hence produces higher accuracy.