Currency Exchange Forecasting Using Sample Mean Estimator and Multiple Linear Regression Machine Learning Models

Abstract

In recent time, there is an increasing growth in the amount of trading taking place in the currency exchange market. However, effective analysis and simulation tools for performing accurate prediction of these exchange rates are lacking. To alleviate this challenge, this work presents a hybrid machine learning and prediction model by suitably combining the Sample Mean Estimator (SME) simulation architecture with the multiple linear regression technique-based training of feed-forward parameters. The developed model has the capability to overcome prediction inaccuracy, inconsistent forecasting, slow response due to computational complexity and scalability problems. The SME method is used to overcome the problems of uncertainty and non-linearity nature of the predictive variable as it’s always affected by economic and political factors. The implementation of the proposed currency exchange rate forecasting system is achieved through the use of a developed in-house Java program with Net Beans as the editor and compiler. Performance comparison between the present system and two baseline methods which are the Autoregressive Moving Average and the Deep Belief network techniques demonstrates that the present forecasting model out-performed the baseline methods studied. The experimental result shows that the precision rate of the present system is equal to or greater than 70%. Therefore, the present foreign exchange predictive system is capable of providing usable, consistent, efficient, faster and accurate prediction to the users consistently at any-time.