Customer churn — sample TabICL report

TabICL classification · target: churned · 300 rows × 6 columns · generated 2026-07-31 19:57

Model quality

Metrics come from a held-out split the model never saw during fitting (80/20 train/test).

0.85

Accuracy

share of correct predictions

0.8373

Balanced accuracy

accuracy averaged over classes

0.84

F1 (macro)

balance of precision and recall

0.8496

ROC AUC

0.5 = random, 1.0 = perfect

Rows are the true classes, columns the predictions. Blue diagonal = correct; red off-diagonal = mistakes.

	pred: no	pred: yes
true: no	33	4
true: yes	5	18

no		186 (62%)
yes		114 (38%)

Permutation importance: how much model quality drops when a column's values are shuffled. Near zero = barely used.

Generated by TabICL MCP — in-context tabular ML (TabICL, Inria).