Predictive Modeling in Forex Trading: A Time Series Analysis Approach

Abstract

This research explores the application of predictive modeling for time series analysis in the context of Foreign Exchange (FOREX) trading, focusing specifically on the USD, EUR, and GBP currency pairs against BDT. The study endeavors to develop a neural network model for accurate exchange rate predictions, aiming to address the challenges posed by uncertainties inherent in the FOREX market. By utilizing a dataset spanning 15 years with 11,745 data points, the research employs feature engineering techniques, including min-max normalization, to enhance model performance. Various machine learning algorithms such as Support Vector Regression, Random Forest Regressor, Decision Tree Regressor, K-neighbors Regressor, SGD Regressor, and XGBRegressor, as well as deep learning algorithms including Convolutional Neural Network, Long Short-Term Memory, Bidirectional LSTM, and several hybrid models, are explored in predicting closing prices for selected currency pairs. The study reveals the effectiveness of these models, with the Random Forest Regressor and bidirectional LSTM standing out among the respective categories.