---
title: RamanBench
emoji: 🥇
colorFrom: green
colorTo: purple
sdk: gradio
app_file: app.py
pinned: true
license: cc-by-4.0
short_description: RamanBench
sdk_version: 6.12.0
tags:
- leaderboard
---

# RamanBench Leaderboard

**RamanBench** is a large-scale benchmark for machine learning on Raman spectroscopy,
unifying **74 publicly available datasets** (163 prediction targets) across four application domains —
Material Science, Biological, Medical, and Chemical — covering both classification and regression tasks.

We compare **28 models**, from the domain standard PLS to modern tabular foundation models
(TabPFN, TabICL) and time-series classifiers (ROCKET, Arsenal).

## Links

- 📦 Package: [github.com/ml-lab-htw/RamanBench](https://github.com/ml-lab-htw/RamanBench) — `pip install raman-bench`
- 🗂️ Datasets: [github.com/ml-lab-htw/raman_data](https://github.com/ml-lab-htw/raman_data) — `pip install raman-data`
- 📄 Paper: [arXiv:2605.02003](https://arxiv.org/abs/2605.02003)
- ✏️ Suggest a dataset: [PROPOSED_DATASETS.md](https://github.com/ml-lab-htw/RamanBench/blob/main/PROPOSED_DATASETS.md) — community queue for the next benchmark release

## Citation

```bibtex
@article{koddenbrock2026ramanbench,
  title={RamanBench: A Large-Scale Benchmark for Machine Learning on Raman Spectroscopy},
  author={Koddenbrock, Mario and Lange, Christoph and Legner, Robin and J{\"a}ger, Martin and K{\"o}gler, Martin and Bournazou, Mariano N Cruz and Neubauer, Peter and Biessmann, Felix and Rodner, Erik},
  journal={arXiv preprint arXiv:2605.02003},
  year={2026}
}
```