Time Series Analysis on Forex Exchange using Artificial Neural Network and K-Nearest Neighbor Algorithms

Abstract

Accurate forex rate prediction is crucial for investors and businesses looking to optimize returns and manage their investment portfolios sensibly in the ever-changing financial markets. With its frequent fluctuations, interdependence, and the influence of world economic events, the forex market presents certain inherent difficulties that are the subject of this study. Given these difficulties, the study focuses on developing and application of reliable forecasting models and approaches. The methodology involves the construction and comparison of Artificial Neural Network (ANN) and K-Nearest Neighbor (KNN) models, with the objective of selecting the most accurate model for subsequent multi-step ahead predictions. Results indicate that KNN models outperform ANN models and are chosen for making 30-day ahead predictions. For the short term, KNN's forecasts are valid, but there are issues when looking farther than a week ahead. The study's conclusion highlights how challenging long-term forecasting in the forex market is by its own nature of chaotic and noisy. The results highlight how crucial it is to keep up research and innovate in order to improve forecasting models in this complex financial environment.