How do I depict an imbalanced class?

Add a small bar to the right of each row showing the support (total samples). The visual width contrast highlights imbalance without breaking the heatmap.

PromptsCharts

ChartsText → ImageLow creditEN

Confusion Matrix Heatmap

Square heatmap showing predicted-vs-true counts with row-normalised colors and value annotations.

When to use this prompt

For multi-class classification results in vision, NLP, or medical-imaging papers.

The prompt

A row-normalised confusion matrix heatmap for a 5-class classification task.

Rows (true class, top to bottom): Class A, Class B, Class C, Class D, Class E.
Columns (predicted class, left to right): same.

Cell values (row-normalised percentages, summing to 100% per row):
- A: 92, 3, 2, 2, 1
- B: 4, 88, 5, 2, 1
- C: 2, 4, 90, 3, 1
- D: 1, 3, 4, 86, 6
- E: 1, 2, 2, 9, 86

Colormap: light-to-dark navy, with all values in white text. Diagonal cells (correct predictions) outlined with a thin gold border.

Right side: a small color-scale legend (0% to 100%).

Below the matrix: per-class precision and recall in a small text annotation row.

Style: clean academic heatmap, square cells, sans-serif labels, white background. Optimised for medical / multi-class classification figures.

Variations

Counts (not normalised)

Show absolute counts instead of row-normalised percentages, with a single global colormap. Useful when class imbalance is part of the story.

Tips

Always note whether values are normalised by row, column, or absolute counts. Each tells a different story.
Outline diagonal cells in a contrasting color so readers see correct predictions instantly.
Add per-class metrics below the matrix — readers always ask "what about class C?".