Enhancing Forex Trading Predictions with Machine Learning: Cloud and Local Performance Evaluation

Abstract

This study investigates the performance of five machine learning algorithms-Linear Regression, Decision Tree, Random Forest, Gradient Boosting, and XGBoost-in the context of Forex market prediction. We compare these algorithms in two computational environments: a local machine using scikit-learn (sklearn) and a Spark cluster. The evaluation metrics include training and prediction times, Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R2). Our results reveal that while the Spark cluster generally offers superior computational efficiency, particularly for complex algorithms, the local machine performs better in terms of predictive accuracy for certain models. Linear Regression consistently shows strong performance across both environments, highlighting its suitability for simpler tasks. This research provides valuable insights into the tradeoffs between computational efficiency and predictive accuracy, guiding the selection of appropriate models and environments for large-scale financial forecasting.