machinelearning.js.org

A JavaScript reimagining of Weka — machine learning algorithms that run entirely in your browser. Nothing to install. Everything runs locally.

Features

• Preprocess — Upload CSV or ARFF datasets. Inspect attribute statistics, distributions, and missing values.
• Classify — Run k-NN and Naïve Bayes classifiers. Evaluate with cross-validation or a percentage hold-out split.
• Cluster — Discover structure with k-Means. Visualize cluster assignments and within-cluster variance.
• Visualize — Interactive 2-D scatter plots — select any two attributes and colour by class.

Algorithms

Algorithm	Type	Implementation
k-Nearest Neighbor	Classification	lib/algorithms/knn.js
Naïve Bayes	Classification	lib/algorithms/naiveBayes.js
k-Means	Clustering	lib/algorithms/kmeans.js

Data Formats

• ARFF (Weka's Attribute-Relation File Format) — sample datasets in public/datasets/
• CSV — first row treated as header; last column used as class attribute

Running locally

npm install
npm run dev      # starts on http://localhost:3200

Build and deploy

How it works

The site is a Next.js static export. next build writes the fully-rendered HTML/CSS/JS to out/. That directory is what gets published to GitHub Pages — the master branch is source-only; the live site is served from the gh-pages branch.

master       ← all source code lives here
gh-pages     ← only the contents of out/ live here (managed automatically)

Publish to production

# One command — builds the static export then pushes out/ to gh-pages:
npm run deploy

That runs next build (writes to out/) then uses the gh-pages package to force-push out/ to the gh-pages branch on origin with a single commit.

GitHub Pages is configured (in the repo Settings → Pages) to serve from the gh-pages branch, root /. The custom domain is wired up via public/CNAME (machinelearning.js.org) which is copied into out/ automatically during the build. public/.nojekyll is also copied, which tells GitHub Pages not to run Jekyll — required so the _next/ static assets folder is served correctly.

If you only want to build (no deploy)

npm run export   # produces out/ without touching the gh-pages branch

Useful for inspecting the static output locally:

npm run export
npx serve out   # or: python3 -m http.server 8080 -d out

Committing source changes

Source lives on master. Normal workflow:

git add -p                          # stage changes
git commit -m "feat: description"
git push origin master              # push source
npm run deploy                      # separately push the built site to gh-pages

git push origin master and npm run deploy are independent — the deploy step only touches the gh-pages branch.

Checking the live site

After npm run deploy completes (takes ~30 s for GitHub to propagate):

https://machinelearning.js.org

GitHub Actions is not used — the gh-pages npm package handles the push directly from your local machine.

Inspired by

Witten, Frank, Hall & Pal — Data Mining: Practical Machine Learning Tools and Techniques (4th ed.) University of Waikato

Tech

Next.js 16 · Material UI v6 · Recharts · Material React Table · App Router · Static Export