# Arima

> 使用 ARIMA 进行单变量时间序列预测（forecast）

- Skill: `zafer-liu/arima` (Agent Skill)
- Install (CLI): `npx skillmds add zafer-liu/arima`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zafer-liu/arima/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: Zafer-Liu (https://skillmd.com/u/zafer-liu)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/zafer-liu/arima

---

# ARIMA 预测

确认时间列、频率、目标和预测区间，按时间排序并处理缺口。检查平稳性，选择合理参数，使用时间切分验证，输出预测值、区间和误差，并说明外部冲击限制。

## Tool routing

1. Use `get_schema` to identify the time column, target column, table name, and available covariates.
2. Use `query_data` only to verify ordering, missing timestamps, frequency, and enough rows for modeling.
3. Use `run_analysis` with `analysis_name="Time_Series_ARIMA"` for the actual forecast computation.
4. Use `generate_chart` on the returned forecast or diagnostic result tables after `run_analysis` succeeds.

## Implementation reference

- Tool entry: `agent/tools/business/data.py::_tool_run_analysis`
- Analysis registry: `Function/Analyze/registry.py`
- Analysis implementation: `Function/Analyze/Time_Series_ARIMA/analyze.py`
- Chart implementation: `Function/Charts_generation/chart_generate.py`