FORECASTING CURRENCY EXCHANGE RATES USING ARIMA, ETS AND RNN: A MACHINE LEARNING PERSPECTIVE

Time series forecasting plays a key role in financial decision-making, particularly in predicting exchange rate fluctuations. In this paper we implemented three different methods for forecasting the INR/USD exchange rate: The Autoregressive Integrated Moving Average (ARIMA) method, the Exponential Smoothing (ETS) method, and a Long Short-Term Memory (LSTM)-based Recurrent Neural Network (RNN) approach. The ARIMA (5, 1, and 0) model captures past dependencies using an autoregressive framework, while the ETS model adjusts for trend-based variations using exponential smoothing techniques. The LSTM-based RNN, designed to capture non linear patterns, is trained using MinMax-scaled data with a sequence-based structure. When comparing all 3 models in terms of predictive accuracy, the deep learning models capturing short-term fluctuations and nonlinear trends more effectively. And performance evaluation using Mean Squared Error (MSE) as 36.424298, 17.956008, 0.521610, Mean Absolute Error (MAE) 5.333123, 3.766352, 0.49238, and Root Mean Squared Error (RMSE) as 6.035255, 4.237453, and 0.64872confirms the superior forecasting capability of the deep learning-based approach. These findings highlight the potential of RNNs in financial time series forecasting, offering a robust alternative to traditional statistical models.