ABSTRACT

This study presents the development and evaluation of The Hybrid LSTMGenerative AI ECO-Model for forecasting financial and economic indicators such as EUR/USD exchange rate, utilizing a combination of Long Short-Term Memory (LSTM) networks and generative AI models (GPT-2 and Llama- 3.2-1B). The primary objective was to achieve high prediction accuracy while minimizing computational resource consumption and ensuring ease of use of the model on various devices. The model was trained and tested on historical financial and economic data, including exchange rates, macroeconomic indicators, commodity prices, and sentiment analysis of financial news. Our findings indicate that traditional LSTM models outperform generative AI models in time-series forecasting tasks due to their ability to capture temporal dependencies. However, integrating generative AI for dataset refinement and model optimization significantly improved forecasting performance. The hybrid ECO-model, leveraging generative AI-driven parameter selection and sentiment analysis, demonstrated superior accuracy for long-term predictions. The most influential parameters included historical exchange rate trends, gold and oil prices and news sentiment. By implementing an ECO-approach that optimizes dataset size, minimizes training iterations, and employs a lightweight model architecture, our study highlights a path toward efficient and sustainable financial forecasting. Future research directions include enhancing anomaly detection mechanisms, incorporating additional weak predictors, and refining the role of generative AI in hybrid time-series forecasting models.