Finance & Banking Finance & Banking use casesNACE K64.19Binary ClassificationTarget: credit_risk_classificationOpenML #31 ↗

Credit G

The German Credit (Credit G) dataset contains 1,000 retail banking records across 20 financial, socio-demographic, and credit history attributes designed to predict individual loan default risk ('good' vs. 'bad' credit).

✨ Try in Carla Install Chrome Extension ↗100% In-Browser WebGPU • Zero Cloud Upload

1,000Records (Rows)

20Predictive Features

7 / 14Numeric / Categorical

0.0%Missing Value Ratio

0.20161 - AUC Error

1.41sWebGPU Mean Split

BUSINESS CONTEXT & OBJECTIVEConsumer Lending & Retail Banking (K64.19)

Business Objective: Credit G

01Business Context

Retail banks and consumer credit institutions evaluate loan applications to allocate personal loans, automotive financing, and consumer credit while minimizing balance sheet risk and non-performing loans (NPLs).

02Analytical Objective

Classify credit applicants into 'good' (creditworthy) or 'bad' (high risk of default) risk profiles prior to capital disbursement.

03Economic & Decision Impact

Optimizing credit scoring delivers immediate capital preservation and higher loan acceptance rates. Asymmetric risk weighting is critical: misclassifying a defaulting customer as good incurs 5x the loss of erroneously rejecting a creditworthy applicant (loss of interest margin vs. full principal loss).

ML Benchmark Narrative

On this classic 1,000-row benchmark, traditional gradient-boosted trees (XGBoost, LightGBM, CatBoost) achieve AUC-ROC scores between 0.77 and 0.79 with extensive feature encoding and cost-matrix tuning. In-context tabular foundation models (TabICL / Carla) match or outperform these tuned baselines zero-shot directly within the context window, eliminating manual one-hot encoding for the 14 categorical features.

Source Origin:UCI Machine Learning Repository / Statlog Project / OpenML

Creator:Dr. Hans Hofmann (University of Hamburg) (1994)

License:Open Data Commons Public Domain Dedication (CC0 / Public Domain)

Reference:Paper / Repo ↗

FEATURE SPECIFICATION & DATA DICTIONARY

Target & Feature Column Definitions

Exhaustive business definitions, measurement units, roles, and target variables across all 21 columns.

Col	Variable / Feature Name	Role	Type	Unit / Scale	Description & Business Meaning
A	`credit_risk_classification`Credit Risk Outcome	TARGET	CAT	—	Binary creditworthiness evaluation: 'good' indicates timely repayment with no default; 'bad' indicates default or critical repayment delinquency.
B	`checking_account_status`Checking Account Status	Feature	CAT	DEM	Status of the applicant's existing checking account balance in Deutsche Mark (e.g., '<0 DM', '0<=X<200 DM', '>=200 DM', or 'no checking account').
C	`loan_duration_months`Loan Duration	Feature	NUM	months	Total requested duration and repayment period of the credit contract in months.
D	`credit_history_status`Credit History	Feature	CAT	—	Historical repayment behavior across past loans (e.g., critical/other credits existing, all loans paid back duly, existing loans paid back duly till now, delays in past payments).
E	`loan_purpose`Loan Purpose	Feature	CAT	—	Stated commercial or personal purpose for borrowing funds (e.g., new car, used car, furniture/equipment, radio/television, domestic appliances, repairs, education, business).
F	`loan_amount_dem`Credit Amount	Feature	NUM	DEM	Total principal credit amount requested and disbursed.
G	`savings_balance_bracket`Savings Account Balance	Feature	CAT	DEM	Categorized tier of liquid savings balance or bond holdings (e.g., '<100 DM', '100<=X<500 DM', '500<=X<1000 DM', '>=1000 DM', or 'no known savings').
H	`employment_duration_years`Employment Tenure	Feature	CAT	years	Duration of continuous employment with current employer (e.g., 'unemployed', '<1 year', '1<=X<4 years', '4<=X<7 years', '>=7 years').
I	`installment_rate_pct_income`Installment Rate (% Disposable Income)	Feature	NUM	%	Monthly loan installment obligation represented as a percentage of applicant disposable net income.
J	`marital_status_and_gender`Personal Status & Sex	Feature	CAT	—	Combined demographic attribute representing gender and marital/custody status (e.g., 'male single', 'female div/dep/mar', 'male mar/wid', 'male div/sep').
K	`guarantor_co_debtor_status`Other Debtors / Guarantors	Feature	CAT	—	Indicates whether additional parties are liable for the debt facility (e.g., 'none', 'co-applicant', 'guarantor').
L	`residence_duration_years`Years at Current Residence	Feature	NUM	years	Number of years the borrower has continuously lived at their current residential address.
M	`most_valuable_asset_type`Collateral Property Magnitude	Feature	CAT	—	Highest-value asset category owned by the borrower (e.g., 'real estate', 'building society savings / life insurance', 'car or other', 'unknown / no property').
N	`borrower_age_years`Age	Feature	NUM	years	Age of the primary credit applicant in chronological years.
O	`other_installment_plans`Other Payment Plans	Feature	CAT	—	Existing installment obligations held with competing financial institutions or retail stores (e.g., 'bank', 'stores', 'none').
P	`housing_tenure_type`Housing Type	Feature	CAT	—	Applicant living and property arrangement (e.g., 'rent', 'own', 'for free').
Q	`num_existing_credits_at_bank`Existing Bank Credits	Feature	NUM	count	Number of active credit lines and loans already registered at this specific bank.
R	`employment_qualification_level`Employment Qualification Level	Feature	CAT	—	Professional employment classification (e.g., 'unemployed/unskilled non-resident', 'unskilled resident', 'skilled employee/official', 'management/self-employed/highly qualified employee').
S	`num_maintenance_liable_dependents`Number of Dependents	Feature	NUM	count	Number of individuals financially dependent on the applicant for maintenance.
T	`registered_telephone_flag`Telephone Registered	Feature	CAT	—	Indicates whether a landline telephone is registered in the applicant's name ('yes' vs. 'none').
U	`is_foreign_worker`Foreign Worker Status	Feature	CAT	—	Indicator of whether the applicant is classified as a foreign worker ('yes' vs. 'no').

DATASET PREVIEW

Interactive Data Table

Explore rows, feature values, and target labels for Credit G.

Download ParquetHF Viewer ↗

LOCAL EXECUTION TELEMETRY

TabICLv2 WebGPU Benchmark Results

Complete performance metrics on persisted IID or non-IID evaluation splits, executed 100% locally in the browser sandbox.

Mean Split Latency

1.41s30 splits • 8 ensembles/split

Total Evaluation Runtime

42.41sIncludes warmup & sync

Evaluation Protocol

10×3 SplitsRepeated IID • 30 total

TabICLv2 Score

0.20161 - AUC Error

Evaluation Split	Train Rows	Test Rows	Split Duration?Wall-clock inference time taken for this evaluation split running 100% locally via WebGPU.Learn more →	Accuracy?Proportion of correct test predictions across all classes.Google ML Guide & Details →	ROC-AUC?Area under the ROC curve evaluating ranking capability across all thresholds.Google ML Guide & Details →	F1 Score?Harmonic mean of Precision and Recall, robust against class imbalance.Learn more →	Precision?Proportion of predicted positives that were actually positive.Google ML Guide & Details →	Recall?Proportion of actual positive cases successfully captured.Google ML Guide & Details →
R1 / F1	666	334	1.46s	77.84%	0.8554	0.8514	0.8030	0.9060
R1 / F2	667	333	1.38s	73.57%	0.7726	0.8204	0.7821	0.8627
R1 / F3	667	333	1.38s	78.68%	0.8232	0.8588	0.8000	0.9270
R2 / F1	666	334	1.39s	74.55%	0.7431	0.8283	0.7854	0.8761
R2 / F2	667	333	1.41s	76.88%	0.8179	0.8438	0.8000	0.8927
R2 / F3	667	333	1.42s	76.58%	0.8306	0.8440	0.7903	0.9056
R3 / F1	666	334	1.39s	75.45%	0.7734	0.8327	0.7969	0.8718
R3 / F2	667	333	1.40s	75.08%	0.7958	0.8350	0.7778	0.9013
R3 / F3	667	333	1.40s	77.18%	0.8170	0.8486	0.7918	0.9142
R4 / F1	666	334	1.41s	77.54%	0.7895	0.8503	0.7978	0.9103
R4 / F2	667	333	1.38s	78.08%	0.8137	0.8549	0.7963	0.9227
R4 / F3	667	333	1.39s	76.28%	0.8062	0.8410	0.7917	0.8970
R5 / F1	666	334	1.40s	74.55%	0.7723	0.8317	0.7749	0.8974
R5 / F2	667	333	1.41s	74.77%	0.8047	0.8320	0.7790	0.8927
R5 / F3	667	333	1.44s	76.58%	0.7848	0.8415	0.7992	0.8884
R6 / F1	666	334	1.39s	73.05%	0.7627	0.8185	0.7748	0.8675
R6 / F2	667	333	1.41s	75.98%	0.7782	0.8374	0.7954	0.8841
R6 / F3	667	333	1.44s	74.47%	0.7974	0.8323	0.7701	0.9056
R7 / F1	666	334	1.44s	75.45%	0.8023	0.8340	0.7923	0.8803
R7 / F2	667	333	1.41s	75.38%	0.7900	0.8299	0.8032	0.8584
R7 / F3	667	333	1.42s	77.48%	0.8002	0.8509	0.7926	0.9185
R8 / F1	666	334	1.39s	78.14%	0.8129	0.8537	0.8038	0.9103
R8 / F2	667	333	1.42s	78.68%	0.8186	0.8560	0.8115	0.9056
R8 / F3	667	333	1.39s	76.88%	0.7832	0.8444	0.7977	0.8970
R9 / F1	666	334	1.41s	79.34%	0.8083	0.8623	0.8090	0.9231
R9 / F2	667	333	1.41s	74.77%	0.7846	0.8243	0.8041	0.8455
R9 / F3	667	333	1.40s	76.88%	0.7994	0.8481	0.7847	0.9227
R10 / F1	666	334	1.45s	76.35%	0.7782	0.8378	0.8063	0.8718
R10 / F2	667	333	1.42s	78.08%	0.8073	0.8554	0.7941	0.9270
R10 / F3	667	333	1.52s	77.48%	0.8256	0.8503	0.7948	0.9142
Mean ± Std	—	—	1.41s	76.40% ± 1.58%	0.798 ± 0.023	0.842 ± 0.012	0.793	0.897

UNIFIED BENCHMARK LEADERBOARD

Foundation Models vs Traditional ML Baselines

Side-by-side evaluation of our in-browser Carla engine (WebGPU), open-source tabular foundation models, and OpenML baselines.

Rank	Algorithm / Model	Model Family	Runtime	Predictive Accuracy?Proportion of correct test predictions across cross-validation splits.Google ML Guide & Details →	ROC-AUC?Area under the ROC curve evaluating ranking capability.Learn more →	F-Measure?Harmonic mean of precision and recall.Learn more →	Reference / Repo
#1	Google TabFM v1.0	Tabular Foundation Model	PyTorch/Python	76.92%	0.8062	0.8439	google-research/tabfm ↗
#2	RandomForest	Random Forest	Python	76.90%	0.7781	0.7490	OpenML #573537 ↗
#3	EXAONE Tabular	Tabular Foundation Model	PyTorch/Python	76.60%	0.8036	0.8424	LGAI-Research/EXAONE-Tabular ↗
#4	Streaming nanotabicl	Tabular Foundation Model	PyTorch/Python	76.50%	0.7984	0.8424	soda-inria/nanotabicl ↗
#5	Full TabICLv2 (PyTorch Reference)	Tabular Foundation Model	PyTorch/Python	76.50%	0.7984	0.8424	soda-inria/tabicl ↗
#6	Carla Engine (TabICLv2 WebGPU)	Tabular Foundation Model	Browser	76.40%	0.7983	0.8417	100% In-Browser
#7	nanotabicl Vanilla	Tabular Foundation Model	PyTorch/Python	76.36%	0.7980	0.8414	soda-inria/nanotabicl ↗
#8	TabPFN v3	Tabular Foundation Model	PyTorch/Python	76.04%	0.7969	0.8385	PriorLabs/tabpfn ↗
#9	NaiveBayes	Naive Bayes	Python	75.30%	0.7815	0.7452	OpenML #278 ↗
#10	AttributeSelectedClassifier GainRatioAttributeEval Ranker NaiveBayes	Naive Bayes	Python	75.30%	0.7815	0.7452	OpenML #565955 ↗
#11	NaiveBayesUpdateable	Naive Bayes	Python	75.30%	0.7815	0.7452	OpenML #573619 ↗
#12	SMO PolyKernel	Support Vector Machine (SVM)	Python	75.10%	0.6688	0.7399	OpenML #338 ↗
#13	SMO PolyKernel	Support Vector Machine (SVM)	Python	75.10%	0.6688	0.7399	OpenML #550232 ↗
#14	AttributeSelectedClassifier GainRatioAttributeEval Ranker SMO	Support Vector Machine (SVM)	Python	75.10%	0.6688	0.7399	OpenML #565802 ↗
#15	RandomRules	Machine Learning Model	Python	74.60%	0.6395	0.7177	OpenML #54760 ↗
#16	AttributeSelectedClassifier A1DE	Machine Learning Model	Python	74.40%	0.7462	0.7226	OpenML #575905 ↗
#17	AttributeSelectedClassifier Bagging REPTree	Decision Tree	Python	74.20%	0.7512	0.7262	OpenML #575630 ↗

STATISTICAL DISTRIBUTIONS

Variable Schema & Summary Distributions

Observed numerical ranges, category cardinalities, missing rates, and sample values across 1,000 rows.

Col	Variable Name	Type	Distinct	Summary Stats / Distribution	Sample Values
A	`credit_risk_classification`TARGET	CAT	2	2 cats: good, bad	`goodbadgood`
B	`checking_account_status`	CAT	4	4 cats: no checking, <0, 0<=X<200 +1 more	`<00<=X<200no checking`
C	`loan_duration_months`	NUM	33	[`4`, `72`] μ=20.9 σ=12.1	`64812`
D	`credit_history_status`	CAT	5	5 cats: existing paid, critical/other existing credit, delayed previously +2 more	`critical/other existing creditexisting paidcritical/other existing credit`
E	`loan_purpose`	CAT	10	10 cats: radio/tv, new car, furniture/equipment +7 more	`radio/tvradio/tveducation`
F	`loan_amount_dem`	NUM	921	[`250`, `18424`] μ=3271.3 σ=2821.3	`116959512096`
G	`savings_balance_bracket`	CAT	5	5 cats: <100, no known savings, 100<=X<500 +2 more	`no known savings<100<100`
H	`employment_duration_years`	CAT	5	5 cats: 1<=X<4, >=7, 4<=X<7 +2 more	`>=71<=X<44<=X<7`
I	`installment_rate_pct_income`	NUM	4	[`1`, `4`] μ=3.0 σ=1.1	`422`
J	`marital_status_and_gender`	CAT	4	4 cats: male single, female div/dep/mar, male mar/wid +1 more	`male singlefemale div/dep/marmale single`
K	`guarantor_co_debtor_status`	CAT	3	3 cats: none, guarantor, co applicant	`nonenonenone`
L	`residence_duration_years`	NUM	4	[`1`, `4`] μ=2.8 σ=1.1	`423`
M	`most_valuable_asset_type`	CAT	4	4 cats: car, real estate, life insurance +1 more	`real estatereal estatereal estate`
N	`borrower_age_years`	NUM	53	[`19`, `75`] μ=35.5 σ=11.4	`672249`
O	`other_installment_plans`	CAT	3	3 cats: none, bank, stores	`nonenonenone`
P	`housing_tenure_type`	CAT	3	3 cats: own, rent, for free	`ownownown`
Q	`num_existing_credits_at_bank`	NUM	4	[`1`, `4`] μ=1.4 σ=0.6	`211`
R	`employment_qualification_level`	CAT	4	4 cats: skilled, unskilled resident, high qualif/self emp/mgmt +1 more	`skilledskilledunskilled resident`
S	`num_maintenance_liable_dependents`	NUM	2	[`1`, `2`] μ=1.2 σ=0.4	`112`
T	`registered_telephone_flag`	CAT	2	2 cats: none, yes	`yesnonenone`
U	`is_foreign_worker`	CAT	2	2 cats: yes, no	`yesyesyes`

FREQUENTLY ASKED QUESTIONS

Frequently Asked Questions: Credit G

Common questions regarding the Credit G dataset, machine learning task formulations, and in-browser tabular inference.

What is the German Credit (Credit G) dataset used for?

The German Credit dataset is a canonical benchmark for evaluating credit risk classification, cost-sensitive learning algorithms, and fairness-aware machine learning in consumer lending.

Can I run zero-server predictions on this dataset in Google Sheets?

Yes, Carla HQ enables zero-server in-context tabular prediction directly in spreadsheets using TabICL foundation models without building pipelines or training infrastructure.

What machine learning models perform best on the German Credit dataset?

Gradient boosting frameworks (LightGBM, XGBoost, CatBoost) and modern in-context tabular foundation models (TabICL / Carla) achieve leading performance (0.78-0.80 AUC-ROC) on this dataset.

LIVE EVALUATION IN GOOGLE SHEETS

Test TabICLv2 on Credit G Yourself

Open the pre-loaded Google Sheet and let Carla configure the target and task for local, zero-cloud tabular machine learning.

Step 1

Launch Carla

Click once to open the spreadsheet and Carla side panel together.

Step 2

Review the Setup

Carla selects the dataset target and task from this page automatically.

Step 3

Evaluate & Predict

Run predictions and compute metrics with zero server uploads.

✨ Try Credit G in Carla Install Chrome Extension ↗

Provenance & Attribution

https://archive.ics.uci.edu/ml/citation_policy.html

License: PublicData Source: OpenMLOpenML Page: https://www.openml.org/d/31

Raw Data Downloads

Download Parquet OpenML Task ↗Google Sheet ↗

← Previous DatasetCompas Two YearsLegal & Public Safety All Datasets Next Dataset →DiabetesHealthcare & Biomedicine