In an increasingly volatile global economic environment, accurate forecasting of currency exchange rates is critical for investors, policymakers, and multinational corporations. This study presents a comparative analysis of three distinct machine learning approaches—Linear Regression (traditional), Random Forest (tree-based ensemble), and Long Short-Term Memory (LSTM, a deep learning model)—to predict the INR/USD exchange rate. Using daily exchange rate data supplemented by key macroeconomic indicators, we examine the predictive accuracy and robustness of each model across multiple performance metrics, including RMSE, MAE, and directional accuracy. The results reveal significant performance differences, with LSTM outperforming in capturing sequential temporal dependencies, while Random Forest demonstrates strong short-term prediction accuracy through non-linear feature interactions. Linear Regression, while easy to implement, is limited in handling volatility and non-linearity. This paper not only highlights the strengths and limitations of each approach but also provides practical insights into the applicability of machine learning models in currency forecasting. Our findings offer a nuanced understanding of model suitability under varying data conditions, contributing to the growing field of AI-driven financial analytics