Scorer/Accuracy Statistics node shows red ❓ for “Overall” row despite perfect per-class metrics


What I want to achieve

  • Evaluate model performance using per-class and overall accuracy metrics.

What’s happening

  • Confusion matrix shows 0 false positives and 0 false negatives.
  • Per-class metrics (Recall, Precision, F‑measure, Specificity, Accuracy) are all 1.0.
  • However, in the “Overall” row, Accuracy and Cohen’s Kappa appear as red question marks (:question:) — not calculated.

I’ve verified that class names match exactly in both columns:

  • Label (Ground Truth): HEALTI, FAIL_S

  • Prediction: HEALTI, FAIL_S
    Troubleshooting steps I’ve taken

  • Checked for typos or mismatches in class labels.

  • Ensured no missing values in either column.

  • Tried manually setting class names in the Scorer dialog.

  • Validated data types and used Table Validator to confirm types are correct.

Question

  • Why is the “Overall” row failing to calculate even though per-class metrics are perfect?
  • Is this due to the custom class names (HEALTHY , FAIL_SOON)?

Thanks
Snehal

Welcome to the KNIME Forum, Snehal_Balghare!
Great to have you here — looking forward to your contributions and discussions.
The model shows perfect scores across all metrics, which is quite rare with real-world data. It might be worth checking for possible data leakage, train/test split issues, or label contamination. If you could share the dataset and your KNIME workflow, we’d be happy to take a closer look together.
Best,
Alpay

Hi Alpay,

Thank you for the warm welcome and your helpful suggestion!

You’re absolutely right — such perfect scores raised a red flag for me too. I’ll definitely double-check for any possible data leakage, train/test split overlap, or label contamination as you mentioned.

In the meantime, I’ll prepare a simplified version of the dataset along with the relevant portion of my KNIME workflow and share it here for review. I really appreciate your offer to take a closer look — that would be a huge help!

Thanks again,
Snehal
train_FD001.txt (3.4 MB)

1 Like