Research map / Label-efficient and robust learning / Recognition and long-tail
A benchmark of categorical encoders for binary classification
Federico Matteucci, Vadim Arzamasov, Klemens Böhm
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use One-Hot, Sum, Binary, or Weight of Evidence rather than other encoders | Binary classification with all evaluated ML models · 50 binary classification datasets | average rank | Improved |
| Use One-Hot rather than other encoders | Decision trees · Binary classification datasets | average rank | Mixed |
| Use Sum, One-Hot, Binary, or Weight of Evidence rather than other encoders | Logistic regression · Binary classification datasets | average rank | Improved |
| Encode categorical attributes rather than dropping them | Binary classification pipelines · 50 binary classification datasets | performance | Improved |
| Increase the number of datasets used in the benchmark | Encoder consensus rankings · Disjoint dataset samples | Spearman ranking replicability | Improved |
| Use logistic regression rather than decision trees as the benchmark model | Encoder benchmark · Disjoint dataset samples | ranking replicability | Improved |
| Change aggregation from mean quality to median quality | Encoder consensus ranking · Binary classification datasets | Spearman correlation between consensus rankings: 0.8 | Mixed |
| Change the ML model from decision trees to SVM | Encoder ranking · Binary classification datasets | Spearman correlation between rankings: 0.3 | Mixed |
| Use full pipeline tuning instead of no tuning | ML model-encoder pipeline · Binary classification | pipeline quality | Improved |
| Use full pipeline tuning instead of model tuning | ML model-encoder pipeline · Binary classification | pipeline quality | Improved |
| Use Min-Hash rather than One-Hot | Encoder for binary classification · Medium-sized tabular datasets | performance | No clear effect |
| Use Binary rather than One-Hot or Mean-Target | Logistic regression · Bank marketing | performance | Hurt |
Closest papers by meaning
- PU-BENCH: A UNIFIED BENCHMARK FOR RIGOROUS AND REPRODUCIBLE PU LEARNING (ICLR 2026 · similarity 0.86)
- Learn to Categorize or Categorize to Learn? Self-Coding for Generalized Category Discovery (NeurIPS 2023 · similarity 0.85)
- Feature Coding in the Era of Large Models: Dataset, Test Conditions, and Benchmark (ICCV 2025 · similarity 0.85)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.84)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.83)
- Benchmarking Predictive Coding Networks -- Made Simple (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
