timeseries-scientist-agentic-forecasting

name	timeseries-scientist-agentic-forecasting
title	TimeSeriesScientist: Autonomous Time Series Forecasting via Multi-Agent Reasoning
version	0.0.2
engine	skillxiv-v0.0.2-claude-opus-4.6
license	MIT
url	https://arxiv.org/abs/2510.01538
keywords	["time-series","agents","forecasting","reasoning","automation"]
description	Automate univariate time series forecasting through a four-agent system orchestrating preprocessing, model selection, validation, and reporting. Use when reducing manual effort in forecasting pipelines and improving reproducibility.

TimeSeriesScientist: Autonomous Time Series Forecasting via Multi-Agent Reasoning

TimeSeriesScientist introduces the first end-to-end agentic framework automating univariate time series forecasting through specialized agents handling preprocessing, model selection, validation, and reporting. The approach achieves 38.2% error reduction versus pure LLM baselines.

Core Architecture

• Four specialized agents: Curator, Planner, Forecaster, Reporter
• Curator agent: Data preprocessing and outlier detection
• Planner agent: Model selection and hyperparameter configuration
• Forecaster agent: Ensemble forecasting with validation
• Reporter agent: Result summarization and uncertainty quantification
• 21 model implementations: Diverse algorithms from statistical to neural

Implementation Steps

Setup multi-agent forecasting system:

# Initialize TimeSeriesScientist framework
from timeseries_scientist import ForecastingMAS, Agent, PreprocessingPipeline

# Create specialized agents
curator = Agent(
    role="data_curator",
    capabilities=["outlier_detection", "missing_value_handling", "detrending", "deseasonalization"],
    model="gpt-4o"
)

planner = Agent(
    role="model_planner",
    capabilities=["model_selection", "hyperparameter_tuning", "ensemble_design"],
    model="gpt-4o"
)

forecaster = Agent(
    role="forecaster",
    capabilities=["forecast_generation", "uncertainty_quantification", "ensemble_combination"],
    model="gpt-4o"
)

reporter = Agent(
    role="report_generator",
    capabilities=["result_summarization", "insight_extraction", "limitation_discussion"],
    model="gpt-4o"
)


mas = ForecastingMAS(
    agents=[curator, planner, forecaster, reporter],
    models_available=,  
    ensemble_strategy=
)