Cross-Validation Score Plot Calculator

Results

Results appear here after calculation.

Enter fold scores, choose settings, then select Calculate Plot.

Interactive score plot

Model statistics

Interpretation

Warnings and diagnostics

Statistical comparison

Fold-level results

Data and configuration

Fold score data

Use CSV, tab-separated, semicolon-separated, or whitespace data. Required columns are Model, Fold, and Validation.

Upload CSV or TSV

Example dataset

Cross-validation method

Expected folds

Expected repeats

Leave-P-Out value

Time-series window

Minimum training window

Test window size

Time-series gap

Group column name

Shuffle folds before splitting

Metric category

Scoring metric

Better score direction

Score display

Decimal precision

Plot type

Custom chart title

Chart height

X-axis label

Y-axis label

Legend position

Marker size

Line width

Bar spacing

Chart theme

Confidence level

Custom confidence percent

Confidence interval method

Bootstrap iterations

Statistical comparison test

Significance level

Multiple-test correction

Outlier method

Generalisation gap threshold

High-variation CV threshold

Training fraction

Random seed

Reference or baseline model

Models to include

Leave every option unselected to include all models.

Minimum score filter

Maximum score filter

Minimum numeric fold

Maximum numeric fold

Show training scores

Show fold points

Show mean lines

Show median lines

Show confidence intervals

Sort models by score

Hide outlier folds

Show data labels

Accepted data format

Use one row for each model, repeat, and fold. Training scores are optional. Validation scores are required.

Model,Repeat,Fold,Train,Validation
Logistic Regression,1,1,0.932,0.901
Logistic Regression,1,2,0.925,0.889
Random Forest,1,1,0.989,0.918

Formula used

Mean score

mean = Σ score / k

Sample standard deviation

SD = √[Σ(score − mean)² / (k − 1)]

Generalisation gap

gap = training score − validation score

Standard error

SE = SD / √k

Coefficient of variation

CV = SD / |mean|

Z-score

z = (score − mean) / SD

How to use

Paste fold scores or upload a CSV file.
Select the metric and score direction.
Choose a plot and confidence method.
Set diagnostics and significance options.
Select Calculate Plot to view results.
Export charts, tables, or the full report.

Example interpretation table

Pattern	Possible meaning	Recommended action
High mean, low variation	Strong and stable performance	Consider final evaluation on held-out data.
High training gap	Possible overfitting	Increase regularisation or simplify the model.
Large fold variation	Unstable validation results	Inspect split quality and dataset size.
One weak fold	Outlier or distribution shift	Review class balance and unusual observations.

Frequently asked questions

What does a cross-validation score plot show?

It shows performance across validation folds. Variation reveals model stability. Comparisons help identify stronger models.

Should higher scores always rank first?

No. Error metrics usually prefer lower values. Select the correct direction before ranking.

Why do fold scores vary?

Each fold contains different observations. Class balance and difficult samples can change performance. Large variation deserves investigation.

What is a generalisation gap?

It is the difference between training and validation performance. Large gaps can suggest overfitting. Metric direction still matters.

Which confidence interval method should I use?

Student's t is useful for small samples. Bootstrap intervals make fewer distribution assumptions. Normal intervals are simple approximations.

Can I compare several models?

Yes. Add each model in the Model column. The calculator ranks and tests available model pairs.

Can this calculator handle repeated cross-validation?

Yes. Use the Repeat column for repeated runs. Each repeat and fold becomes one observation.

How are outliers detected?

The IQR method uses quartile fences. The z-score method uses standard deviations. You can disable outlier detection.

Does statistical significance prove one model is best?

No. Significance describes evidence under assumptions. Practical impact, stability, cost, and test-set performance also matter.