User Guide Overview
The Data Valuation Engine helps users assess data quality, understand model behavior, and identify or mitigate spatial bias on the data relevant to the purpose/problem that is to be solved.
What You Can Do
The application currently supports the following analysis modules:
| Feature | What it does | What you see |
|---|---|---|
Feature Correlation |
Computes pair-wise Pearson and/or Spearman and/or Kendall correlation analysis for all numeric features. Heat-map colouring reveals strength/direction, and can be downloaded as CSV. |
Correlation tables, CSV download buttons. |
Feature Importance |
Trains four tree-based models — Decision Tree, Random Forest, Extra Trees, LightGBM — with 10-fold cross-validation and ranks them by mean accuracy. SHAP post-analysis explains the winning model and displays feature-level scores. |
Model leaderboard, SHAP values table, CSV download button. |
Spatial Bias Audit |
Audits for spatial bias using statistical tests and quantifies it using the Spatial Bias Index (SBI). |
Interactive maps (population distribution, bias heatmaps), bias per region table. |
Spatial Bias Mitigation |
Post-processes model outputs to mitigate spatial bias using either relabeling or regional decision boundary adjustment, optimizing SBI with minimal model change. |
Interactive maps (population, pre/post bias heatmaps, performed flips), pre/post performance metrics and bias tables, adjusted thresholds (for decision boundary adjustment) |
Getting Started
Accessing the UI
By default, the UI runs locally at http://localhost:3000/ and it is online in the AI-DAPT platform at https://data-valuation-ui.aidapt.s5projects.eu/.
The Home page provides quick access to either standalone modules (Correlation Analysis, Feature Importance) or Tasks generated by data pipelines.
Tasks allow users to analyze or interact with datasets as part of automated data pipelines that require user interaction or validation before the pipeline continues. Each task represents a dataset and enables users to apply Spatial Bias Audit and Spatial Bias Mitigation workflows.
| Correlation Analysis and Feature Importance will be supported for tasks in a future release. |
Each of the three modules (Correlation Analysis, Feature Importance, and Spatial Bias Audit or Mitigation) follows a guided workflow: configure → run → inspect results → save outputs.
Dataset Introduction
For Feature Correlation and Feature Importance, users must provide the dataset-id as a URL parameter, for example https://data-valuation-ui.aidapt.s5projects.eu/feature-importance?dataset-id=XXX or https://data-valuation-ui.aidapt.s5projects.eu/correlation?dataset-id=XXX. This allows the Data Valuation Engine to fetch the dataset metadata from the Data Documentation service and download the dataset itself from the Scalable Distributed Storage service.
If no dataset is selected, the UI displays the corresponding error message.