access_approvedOpenML #4135 ↗Amazon Employee Access
Enterprise identity and access management benchmark containing 32,769 historical access requests to automate employee permission approvals and denials based on organizational roles.
Business Objective: Amazon Employee Access
01Business Context
Enterprise employees continuously require specific software, database, and portal permissions as they transition across departments, projects, and seniority levels. Manual supervisor reviews create operational bottlenecks, delay onboarding, and increase IT ticket overhead.
02Analytical Objective
Predict whether an employee's requested access to a specific enterprise resource should be automatically approved (1) or denied (0) based on their manager and organizational role hierarchy.
03Economic & Decision Impact
Automates access provisioning workflows, slashing employee onboarding latency and IT helpdesk costs while mitigating compliance risks from wrongful authorization (false positives) and productivity loss from unnecessary denials (false negatives).
Due to high-cardinality categorical features masked as integer IDs, gradient boosted trees (CatBoost, LightGBM, XGBoost) combined with target encoding and frequency encoding historically dominate this benchmark with AUC-ROC scores exceeding 0.88-0.90. In-context tabular foundation models (TabICL / Carla) achieve competitive zero-shot classification directly on spreadsheet inputs by inferring cross-feature role hierarchies without manual feature engineering pipelines.
Target & Feature Column Definitions
Exhaustive business definitions, measurement units, roles, and target variables across all 10 columns.
| Col | Variable / Feature Name | Role | Type | Unit / Scale | Description & Business Meaning |
|---|---|---|---|---|---|
| A | access_approvedAccess Approval Status | TARGET | NUM | — | Binary decision target indicating whether access to the specified resource was approved (1) or denied/revoked (0). |
| B | resource_idResource ID | Feature | NUM | — | Anonymized unique integer identifier representing the specific IT resource, database, or software application requested by the employee. |
| C | manager_idManager Employee ID | Feature | NUM | — | Unique anonymized employee identifier of the direct supervisor or manager responsible for the requesting employee. |
| D | role_rollup_1Role Rollup Level 1 | Feature | NUM | — | Highest-level organizational group identifier representing broad corporate divisions (e.g., US Engineering vs. EU Operations). |
| E | role_rollup_2Role Rollup Level 2 | Feature | NUM | — | Secondary organizational subdivision category identifier providing narrower departmental context within Rollup 1. |
| F | role_department_idDepartment ID | Feature | NUM | — | Anonymized integer code corresponding to the employee's specific functional department (e.g., Retail Logistics, Security Infrastructure). |
| G | role_title_idJob Title ID | Feature | NUM | — | Anonymized identifier corresponding to the formal business title of the employee (e.g., Senior Software Development Engineer). |
| H | role_family_description_idRole Family Description ID | Feature | NUM | — | Anonymized identifier capturing the granular role family descriptor and technical track classification. |
| I | role_family_idRole Family ID | Feature | NUM | — | High-level job family group identifier categorizing the broader profession (e.g., Software Engineering, Product Management). |
| J | role_code_idUnique Role Code | Feature | NUM | — | Granular unique code assigned to the exact role profile combining title, level, and departmental hierarchy. |
Interactive Data Table
Explore rows, feature values, and target labels for Amazon Employee Access.
TabICLv2 WebGPU Benchmark Results
Complete performance metrics on persisted IID or non-IID evaluation splits, executed 100% locally in the browser sandbox.
| 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 | 21,846 | 10,923 | 129.26s | 94.93% | 0.8425 | 0.9736 | 0.9561 | 0.9917 |
| R1 / F2 | 21,846 | 10,923 | 129.13s | 95.02% | 0.8560 | 0.9740 | 0.9578 | 0.9908 |
| R1 / F3 | 21,846 | 10,923 | 129.26s | 94.90% | 0.8602 | 0.9734 | 0.9576 | 0.9897 |
| R2 / F1 | 21,846 | 10,923 | 129.19s | 95.03% | 0.8481 | 0.9741 | 0.9577 | 0.9910 |
| R2 / F2 | 21,846 | 10,923 | 129.24s | 95.13% | 0.8500 | 0.9746 | 0.9582 | 0.9915 |
| R2 / F3 | 21,846 | 10,923 | 129.23s | 94.97% | 0.8599 | 0.9738 | 0.9577 | 0.9904 |
| R3 / F1 | 21,846 | 10,923 | 129.22s | 95.00% | 0.8593 | 0.9739 | 0.9580 | 0.9904 |
| R3 / F2 | 21,846 | 10,923 | 129.12s | 94.85% | 0.8488 | 0.9732 | 0.9559 | 0.9911 |
| R3 / F3 | 21,846 | 10,923 | 129.19s | 95.08% | 0.8459 | 0.9744 | 0.9568 | 0.9926 |
| Mean ± Std | — | — | 129.20s | 94.99% ± 0.08% | 0.852 ± 0.006 | 0.974 ± 0.000 | 0.957 | 0.991 |
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 | classif.ranger | Machine Learning Model | Python | 95.24% | 0.6638 | 0.9447 | OpenML #522992 ↗ |
| #2 | Google TabFM v1.0 | Tabular Foundation Model | PyTorch/Python | 95.23% | 0.8727 | 0.9750 | google-research/tabfm ↗ |
| #3 | EXAONE Tabular | Tabular Foundation Model | PyTorch/Python | 95.20% | 0.8768 | 0.9750 | LGAI-Research/EXAONE-Tabular ↗ |
| #4 | mlr.classif.ranger | Machine Learning Model | Python | 95.14% | 0.8660 | 0.9417 | OpenML #1990987 ↗ |
| #5 | TabPFN v3 | Tabular Foundation Model | PyTorch/Python | 95.02% | 0.8458 | 0.9741 | PriorLabs/tabpfn ↗ |
| #6 | nanotabicl Vanilla | Tabular Foundation Model | PyTorch/Python | 95.00% | 0.8513 | 0.9739 | soda-inria/nanotabicl ↗ |
| #7 | Full TabICLv2 (PyTorch Reference) | Tabular Foundation Model | PyTorch/Python | 95.00% | 0.8517 | 0.9739 | soda-inria/tabicl ↗ |
| #8 | Carla Engine (TabICLv2 WebGPU) | Tabular Foundation Model | Browser | 94.99% | 0.8523 | 0.9739 | 100% In-Browser |
| #9 | Streaming nanotabicl | Tabular Foundation Model | PyTorch/Python | 94.70% | 0.8453 | 0.9726 | soda-inria/nanotabicl ↗ |
| #10 | IBk | k-Nearest Neighbors (k-NN) | Python | 94.70% | 0.8429 | 0.9410 | OpenML #569343 ↗ |
| #11 | classif.IBk | k-Nearest Neighbors (k-NN) | Python | 94.63% | 0.7345 | 0.9452 | OpenML #522989 ↗ |
| #12 | pipeline.Pipeline(imputation=openmlstudy14.preprocessing.ConditionalImputer,hotencoding=sklearn.preprocessing.data.OneHotEncoder,variencethreshold=sklearn.feature selection.variance threshold.VarianceThreshold,classifier=sklearn.tree.tree.DecisionTreeClassifier) | Decision Tree | Python | 94.55% | 0.7168 | 0.9435 | OpenML #8858439 ↗ |
| #13 | AttributeSelectedClassifier IBk | k-Nearest Neighbors (k-NN) | Python | 94.44% | 0.8241 | 0.9231 | OpenML #570005 ↗ |
| #14 | classif.ada | Machine Learning Model | Python | 94.35% | 0.6613 | 0.9380 | OpenML #522988 ↗ |
| #15 | JRip | Machine Learning Model | Python | 94.27% | 0.5272 | 0.9197 | OpenML #568945 ↗ |
| #16 | J48 | Decision Tree | Python | 94.21% | 0.4994 | 0.9140 | OpenML #568577 ↗ |
| #17 | REPTree | Decision Tree | Python | 94.21% | 0.4994 | 0.9140 | OpenML #568697 ↗ |
Variable Schema & Summary Distributions
Observed numerical ranges, category cardinalities, missing rates, and sample values across 32,769 rows.
| Col | Variable Name | Type | Missing | Distinct | Summary Stats / Distribution | Sample Values |
|---|---|---|---|---|---|---|
| A | access_approvedTARGET | NUM | 0 | 2 | [0, 1] μ=0.9 σ=0.2 | 111 |
| B | resource_id | NUM | 0 | 7,518 | [0, 312153] μ=42923.9 σ=34173.4 | 393531718336724 |
| C | manager_id | NUM | 0 | 4,243 | [25, 311696] μ=25989.0 σ=35927.5 | 85475154014457 |
| D | role_rollup_1 | NUM | 0 | 128 | [4292, 311178] μ=116952.6 σ=10875.4 | 117961117961118219 |
| E | role_rollup_2 | NUM | 0 | 177 | [23779, 286791] μ=118301.8 σ=4551.5 | 118300118343118220 |
| F | role_department_id | NUM | 0 | 449 | [4674, 286792] μ=118912.8 σ=18961.0 | 123472123125117884 |
| G | role_title_id | NUM | 0 | 343 | [117879, 311867] μ=125916.2 σ=31036.0 | 117905118536117879 |
| H | role_family_description_id | NUM | 0 | 2,358 | [4673, 311867] μ=170178.4 σ=69508.4 | 117906118536267952 |
| I | role_family_id | NUM | 0 | 67 | [3130, 308574] μ=183703.4 σ=100486.9 | 29091930857419721 |
| J | role_code_id | NUM | 0 | 343 | [117880, 270691] μ=119789.4 σ=5784.2 | 117908118539117880 |
Frequently Asked Questions: Amazon Employee Access
Common questions regarding the Amazon Employee Access dataset, machine learning task formulations, and in-browser tabular inference.
What is the Amazon Employee Access dataset used for?
The Amazon Employee Access dataset is a canonical machine learning benchmark containing 32,769 records used to build automated models that predict whether an employee should be granted or denied access to a requested IT resource based on their organizational role attributes.
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 needing Python scripts or server deployment.
What machine learning models perform best on Amazon Employee Access?
Gradient boosting frameworks like CatBoost and LightGBM with target encoding and frequency encoding achieve AUC-ROC scores between 0.88 and 0.91. In-context tabular foundation models such as TabICL deliver immediate zero-shot predictions without manual encoding pipelines.
Test TabICLv2 on Amazon Employee Access Yourself
Open the pre-loaded Google Sheet and let Carla configure the target and task for local, zero-cloud tabular machine learning.
Launch Carla
Click once to open the spreadsheet and Carla side panel together.
Review the Setup
Carla selects the dataset target and task from this page automatically.
Evaluate & Predict
Run predictions and compute metrics with zero server uploads.
Provenance & Attribution
OpenML Dataset 4135