Binary Classification of Exchange Rate Trends Using Logistic Regression

Abstract

This study explores the use of logistic regression for binary classification of exchange rate trends, focusing on predicting upward and downward currency movements. Logistic regression, valued for its simplicity and interpretability, models the relationship between historical exchange rate data and macroeconomic indicators like interest rates, inflation, GDP growth, and trade balances. The methodology involves data collection, preprocessing, feature engineering, and model evaluation. Historical data is processed to address missing values, outliers, and noise, ensuring a robust dataset. Feature selection techniques, including mutual information scores and principal component analysis (PCA), identify key predictors, while L1 and L2 regularization enhance generalization. The model, implemented using Python's scikit-learn library, is optimized through grid search for hyperparameter tuning. Performance metrics, including accuracy, precision, recall, F1-score, and ROC-AUC, indicate strong predictive capability, achieving 99% accuracy in forecasting upward and downward trends. Logistic regression's interpretability aids decision-making, making it a valuable tool for financial forecasting. However, the study notes limitations, such as challenges posed by market volatility and geopolitical factors. Future research suggests incorporating sentiment analysis from financial news and social media, and exploring hybrid models combining logistic regression with ensemble methods or deep learning to improve performance under real-world conditions.