Prompt-Driven Time Series Forecasting with Large Language Models

Zairo Bastos, João David Freitas, José Wellington Franco, Carlos Caminha, Carlos Caminha

2025

Abstract

Time series forecasting with machine learning is critical across various fields, with Ensemble models and Neural Networks commonly used to predict future values. LSTM and Transformers architecture excel in modeling complex patterns, while Random Forest has shown strong performance in univariate time series forecasting. With the advent of Large Language Models (LLMs), new opportunities arise for their application in time series prediction. This study compares the forecasting performance of Gemini 1.5 PRO against Random Forest and LSTM using 40 time series from the Retail and Mobility domains, totaling 65,940 time units, evaluated with SMAPE. Results indicate that Gemini 1.5 PRO outperforms LSTM by approximately 4% in Retail and 6.5% in Mobility, though it underperforms Random Forest by 5.5% in Retail and 1% in Mobility. In addition to this comparative analysis, the article contributes a novel prompt template designed specifically for time series forecasting, providing a practical tool for future research and applications.

Paper Citation

in Harvard Style

Bastos Z., Freitas J., Franco J. and Caminha C. (2025). Prompt-Driven Time Series Forecasting with Large Language Models. In Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS; ISBN 978-989-758-749-8, SciTePress, pages 309-316. DOI: 10.5220/0013363800003929

in Bibtex Style

@conference{iceis25,
author={Zairo Bastos and João Freitas and José Franco and Carlos Caminha},
title={Prompt-Driven Time Series Forecasting with Large Language Models},
booktitle={Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS},
year={2025},
pages={309-316},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013363800003929},
isbn={978-989-758-749-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS
TI - Prompt-Driven Time Series Forecasting with Large Language Models
SN - 978-989-758-749-8
AU - Bastos Z.
AU - Freitas J.
AU - Franco J.
AU - Caminha C.
PY - 2025
SP - 309
EP - 316
DO - 10.5220/0013363800003929
PB - SciTePress