How does stratified splitting handle imbalanced class labels?

It groups rows by the target column, shuffles each class independently using the random seed, and distributes members across splits using largest-remainder allocation so each split matches the original class distribution.

What is the difference between standard split and k-fold cross-validation mode?

Standard split divides data into fixed train, validation, and test sets based on percentage ratios, whereas k-fold splits data into k non-overlapping validation folds with round-robin stratified assignment.

How does the SMOTE preview work?

SMOTE creates synthetic samples exclusively for minority classes in the training split by interpolating between nearest minority neighbors across numeric feature columns.

What happens if a class has fewer samples than the number of splits or folds?

Classes with fewer than two samples trigger rare-class guards, and requesting more folds (k) than the sample count of the smallest class is rejected to prevent invalid cross-validation partitions.

What files are included in the downloaded ZIP archive?

The ZIP export contains partitioned CSV files (e.g., train.csv, validation.csv, test.csv, or fold-specific CSVs) along with a split_report.json manifest detailing distribution stats and configurations.

Elysia Tools

Navigation

Data Analysis

Train/Test Split with Stratification

Class-stratified train/validation/test split or stratified k-fold for CSV/JSON datasets — seeded, reproducible, with distribution reports, leakage checks, SMOTE preview and CSV export.

Details

What this tool helps you do

Implements sklearn-style semantics in pure TypeScript: per-class seeded shuffle (mulberry32 + Fisher–Yates) with largest-remainder allocation across splits (or StratifiedKFold round-robin for k-fold), per-split class-share tables with deviation bars, cross-split duplicate-leakage checks, rare-class guards (classes with fewer than 2 members cannot stratify — sklearn raises the same error; k > smallest class is rejected like sklearn's nsplits error), and a SMOTE preview (Chawla et al. 2002: xnew = xi + λ·(xneighbor − xi) with one λ per synthetic sample, over the 5 nearest minority neighbors, numeric columns only) applied to the training split only. Exports train/validation/test (or per-fold) CSVs plus a splitreport.json manifest in a ZIP.

Execution

Run this tool

Fill in the form, run the tool, and review the result in one place.

Prepared example runs

Click an example to fill the form automatically. File inputs still need an upload.

2 examples

Stratified 70/15/15 split of a small sentiment dataset

24 rows in a 10/9/5 class mix split 16/5/3 with stratification — the report shows each split's class shares against the dataset's with deviation bars, and exports train/validation/test CSVs plus a split_report.json in a ZIP.

<section class="tts">16/5/3 split · 3 classes (10/9/5) · deviation bars per split · train.csv + validation.csv + test.csv + split_report.json in ZIP</section>

Stratified 4-fold cross-validation plan

The same dataset as 4 stratified folds sized 8/6/5/5 — each validation fold keeps the class mix within one sample of proportional, the fold plan block shows the round-robin assignment, and per-fold train/validation CSVs land in the ZIP.

<section class="tts">4 folds (8/6/5/5 rows) · class proportions preserved per fold · fold_i_train.csv + fold_i_validation.csv × 4 + report in ZIP</section>

Inputs

Options are grouped so long tools stay easier to scan.

12 options

FilesUpload source files for this workflow.1

Dataset file (CSV or JSON)fileOptional

Single file max size: 50 MBSupported types: text/csv, application/json, application/octet-stream, .csv, .json

ContentPaste or type the main input values.2

Or paste CSV (with header)textareaOptionalTarget column (class label)textRequired

SettingsAdjust formats, ranges, numbers, and modes.5

Split modeselectOptionalTraining percentnumberOptionalValidation percentnumberOptionalNumber of folds (k)numberOptionalRandom seednumberOptional

TogglesEnable or disable optional behavior.4

Stratify by target columncheckboxOptionalEnabled when checkedShuffle rows before splittingcheckboxOptionalEnabled when checkedSMOTE oversampling preview (train split)checkboxOptionalEnabled when checkedExport split CSVs as ZIPcheckboxOptionalEnabled when checked

Result

Ready for a run

Run the tool to preview files, text, structured data, or streamed output here.

Samples

Train/Test Split with Stratification

What this tool helps you do

Run this tool

Prepared example runs

Inputs

Result

Examples that match this tool

Continue with connected tools and hubs

Prepared example runs

Inputs

Result

Learn when to use this tool, what it supports, and how real users apply it.

Key facts

Overview

When to use

How it works

Use cases

Examples

1. Stratified 70/15/15 Split for Sentiment Classification

2. Stratified 4-Fold Cross-Validation for Tabular Risk Scoring

FAQ

CSV Samples

Python Samples

WebRTC Real-Time Communication Samples

Distributed Tracing Samples

Time Series Forecast & Seasonality Analyzer

CSV to Database Migration Planner

CSV/JSON Data Watermarker

Dataset Imbalance Detector & Resampler

JSON Interchange and Format Translation Tools

Text Case, Encoding, and Normalization Conversion Tools

CSV Export and Table Conversion Tools

JSON Inspection, Diff, and Transformation Tools