LLM4Series: Structured Prompting for Time Series Forecasting with LLMs

Wesley Barbosa Silva, Maria Fernanda Aquino Freitas Scarcela, Luiz Zairo Bastos Viana, Carlos Caminha, João Paulo do Vale Madeiro, José Wellington Franco da Silva

Presentation Attendance: Yes, we will present in-person

Keywords: Time Series Forecasting, Large Language Models, Prompt Engineering, Open-source Library

TL;DR: We introduce LLM4Series, an open-source Python library that standardizes and simplifies time series forecasting with Large Language Models through structured prompting and a unified evaluation pipeline.

Abstract: We introduce LLM4Series, an open-source Python library that standardizes time series forecasting with Large Language Models (LLMs). Despite the growing interest in applying LLMs to temporal tasks, the lack of dedicated tooling hinders systematic experimentation. LLM4Series addresses this gap through a modular and extensible pipeline encompassing data serialization, structured prompt construction, and automated evaluation. We validate the framework on three real-world datasets, comparing it against statistical and deep learning baselines. Results show that our structured methodology achieves competitive accuracy while substantially reducing the complexity of LLM-based forecasting.

Track: Research Track (max 4 pages)

Submission Number: 94