کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Exploring Different Machine Learning-based Methods for Learning the Language of Shepna Stock Price

نویسندگان :

Zoreh Ansari¹ Jalal Raeisi Gahruei² Mansoor Khademi³

1- Esfahan Oil Refinery 2- Esfahan Oil Refinery 3- Esfahan Oil Refinery

کلمات کلیدی :

Time series forecasting،Long Short-Term Memory،Reservoir Computing Neural Network،Large language models،Zero-shot learning

چکیده :

Accurate stock price forecasting is critical for financial decision-making and market policy formation. Despite the growing use of machine learning and deep learning methodologies for stock market prediction, reliably predicting stock prices remains a persistent challenge. In this work, we explore a bioinspired neural architecture, Reservoir Computing (RC), alongside more conventional LSTM-based networks for predicting the Shepna (Esfahan Oil Refinery) stock price. Additionally, we investigate the Chronos model family—an emerging class of Large Language Model (LLM)-based time series forecasting architectures capable of zero-shot learning, thus facilitating transfer learning in time series prediction tasks. Experimental evaluations compare the predictive performance of RC, LSTM, and Chronos under varied hyperparameter configurations. Results indicate that an optimized RC architecture consistently outperforms both LSTM and Chronos, suggesting the robustness of RC for real-world financial applications. We further analyze how key parameters influence Chronos’ zero-shot forecasting capability, providing insight into LLM-driven approaches for time series modeling. These findings underscore the potential benefits of bioinspired neural architectures and large language model-based methods in advancing stock price prediction.