Enhancing Economic Growth Time Series for UAE Forecasting with Deep Learning: A Seq2Seq and Attention-Driven LSTM Approach

Abstract

This study aims to enhance economic growth forecasting in the United Arab Emirates (UAE) by implementing a Seq2Seq deep learning model with an attention-driven Long Short-Term Memory (LSTM) network. Traditional statistical models often fail to capture the complex temporal dependencies and nonlinear trends inherent in economic time series data. To address these limitations, this research employs a structured methodology, beginning with data collection from the World Bank, including macroeconomic indicators such as GDP growth, inflation, trade balance, investment flows, and employment rates. Preprocessing steps involve handling missing values, normalization, and feature engineering. The proposed Seq2Seq LSTM model utilizes an encoder-decoder structure with an attention mechanism to assign dynamic weights to critical time points, improving forecasting accuracy. The model is trained using the Adam optimizer and evaluated using RMSE, MAE, and MAPE metrics. Results demonstrate superior predictive performance compared to traditional approaches, with improved generalization on unseen data. Findings suggest that attention-enhanced deep learning models provide more reliable economic forecasts, aiding policymakers in decision-making. Future work should explore hybrid models, incorporate external economic shocks, and optimize hyperparameter tuning for further accuracy improvements.