High-Accuracy Forex Trading Prediction Model Using Machine Learning Algorithms

Abstract

The foreign exchange (Forex) market, the largest financial market globally, involves the simultaneous buying and selling of currency pairs such as the United States Dollar (USD) and the Japanese Yen (JPY). Due to the high leverage and volatility inherent in Forex trading, retail investors and traders face significant financial risks. To mitigate this, a predictive model was developed using supervised machine learning to improve trading outcomes. A dataset comprising EUR/USD currency pair data from MetaTrader5, covering a period of 7 years (2015–2022), was streamlined to 5 years (2017–2022) for relevance. The dataset contained 13 features and 1,029 records, which underwent data cleaning and processing. Feature selection was performed using the Boruta package in R, and various machine learning algorithms were employed for model training and testing. The model’s accuracy was evaluated using the ROC curve and the Kappa statistic. The inclusion of the Simple Moving Average indicator improved forecast accuracy. Among the algorithms tested, the Generalized Linear Model (Logistic Model) performed best, achieving an accuracy of 86.28%, a specificity of 87.42%, a sensitivity of 86.5%, and an AUC (ROC) score of 86.97%. This predictive model has the potential to optimize profits, particularly in the context of binary options trading within the Forex market.