Tools / Feature-Space Coverage

Feature-Space Coverage Analysis

This tool checks whether your dataset has gaps in the combined feature space. Unlike column-by-column checks, this analysis looks at how features interact: are there combinations of feature values where you have little or no data? These are the regions where your model will be forced to extrapolate, potentially producing unreliable predictions.

How it works: We project your features into a 2D map using PCA, divide the map into a grid, and identify cells with few or no samples. For each gap, we report which original features define that region so you can decide whether to collect more data.

Drop a CSV file here or click to upload

No login required. Raw upload is processed for analysis and not retained by default.

Max 50MB · at least 50 rows · at least 3 numeric columns

If provided, label distribution per region is shown and color-by-label toggle is enabled.

Leave blank to auto-select based on dataset size.

What you get

Coverage score

A single 0–100 score summarising how well your dataset covers the combined feature space, with a colour-coded verdict.

2D coverage map

An interactive heatmap showing dense, sparse, and empty regions. Overlays actual data points and flags gap cells with red outlines.

Flagged gap profiles

For each gap, the exact original-feature ranges that define it — so you know what data to collect, not just where the hole is.

Confidence scoring

Each flagged region is rated High / Medium / Low confidence based on PCA quality, interior-ness, and local neighbourhood consistency.