🌸 Iris Flower Classification — Machine Learning in Python (scikit-learn, FastAPI) with a Live In-Browser Demo
An end-to-end machine learning project in Python: exploratory data analysis of the classic UCI Iris dataset,
comparison of eight scikit-learn classifiers with stratified cross-validation, hyperparameter tuning, evaluation,
interpretation, a FastAPI /predict REST API, Docker deployment, GitHub Actions CI and an interactive web demo
on GitHub Pages where the trained model runs entirely in your browser (JavaScript inference from exported parameters).
Research work carried out as part of the 6-Month Industrial Training & Certification Program in Artificial Intelligence and Machine Learning, jointly offered by IIT Ropar and NIELIT Ropar, under MeitY, Govt. of India.
10-fold stratified cross-validation, default hyperparameters, standard scaling in each pipeline:
| Model | CV accuracy | SD |
|---|---|---|
| LDA | 0.973 | 0.044 |
| K-nearest neighbours | 0.960 | 0.053 |
| SVM (RBF) | 0.960 | 0.044 |
| Logistic regression | 0.953 | 0.052 |
| Random forest | 0.953 | 0.052 |
| Gaussian naive Bayes | 0.953 | 0.052 |
| Decision tree | 0.933 | 0.052 |
| Majority baseline | 0.333 | 0.000 |
Tuning (GridSearchCV, same folds): logistic regression 0.973 (C = 100) · KNN 0.973 (k = 13) · SVM 0.967 (C = 1, γ = 0.1).
Deployed model — tuned logistic regression:
- Out-of-fold accuracy 0.973 (150 flowers); macro F1 0.973. Setosa 1.000 F1; versicolor 0.959; virginica 0.961.
- Stratified 70/30 hold-out: accuracy 0.911 (45 flowers) — a small split, so noisier than CV.
- Most informative features: petal length and petal width (random-forest importances 0.455 and 0.417; largest logistic-regression coefficients).
Browser ↔ Python check: the JavaScript model on the live page reproduces scikit-learn's predict_proba on all 150
flowers with a maximum absolute difference of 4.4 × 10⁻¹⁶ and 100% label agreement.
Details: docs/results.md.
| Desktop | Mobile |
|---|---|
 |
 |
flowchart LR
A[data/iris.data<br/>UCI Iris, CC BY 4.0] --> B[src/train.py<br/>StandardScaler + LogisticRegression<br/>GridSearchCV]
B --> C[models/iris_model.joblib]
B --> D[web/model.json<br/>mean, scale, coef, intercept]
C --> E[app/main.py<br/>FastAPI /predict]
D --> F[web/ app.js<br/>softmax inference in the browser]
F --> G[GitHub Pages live demo]
E --> H[Docker / docker-compose]
notebooks/iris_classification.ipynb EDA, preprocessing, model comparison, tuning, evaluation, interpretation
src/data.py, src/train.py data loading; training + export (joblib and JSON)
app/main.py FastAPI app: GET /health, POST /predict
tests/test_api.py pytest: API, validation, dataset, JSON-export parity with scikit-learn
web/ static demo (index.html, style.css, app.js, model.json, iris.json)
Dockerfile, docker-compose.yml local deployment (API on :8000, demo on :8080)
.github/workflows/ci.yml CI: ruff + pytest on Python 3.12 and 3.13
gh-pages branch published copy of web/ served by GitHub Pages
docs/ extended documentation
git clone https://github.com/Rishabh-bgp/iris-classification.git
cd iris-classification
python -m venv .venv && source .venv/bin/activate
pip install -r requirements-dev.txt
python -m src.train # trains, writes models/ and web/model.json
pytest -q # 6 tests
uvicorn app.main:app --reload # API at http://localhost:8000/docs
python -m http.server -d web 8080 # demo at http://localhost:8080docker compose up --build # API → http://localhost:8000 · demo → http://localhost:8080curl -X POST http://localhost:8000/predict -H "Content-Type: application/json" \
-d '{"sepal_length":5.1,"sepal_width":3.5,"petal_length":1.4,"petal_width":0.2}'Returns {"species": "Iris-setosa", "probabilities": {...}}. Interactive docs at /docs. See docs/api.md.
| Page | Contents |
|---|---|
| docs/data.md | Dataset, attribution, licence |
| docs/methodology.md | Pipeline, models, validation, tuning |
| docs/results.md | Full results and interpretation |
| docs/api.md | REST API reference |
| docs/web-demo.md | How in-browser inference works; design and accessibility |
| docs/deployment.md | Docker, compose, CI, GitHub Pages |
| docs/faq.md | FAQ and limitations |
Code and docs: MIT © 2026 Rishabh Aryan. Iris data: CC BY 4.0 — Fisher, R. A. (1936), UCI Machine Learning Repository, doi:10.24432/C56C76.
| Repository | Description | Live |
|---|---|---|
| 100-days-of-python | My original, offline, no-API-key solutions to all 100 projects of the 100 Days of Code Python bootcamp curriculum… | — |
| Agentic-Engineering-Ebook | Agentic Engineering is a practical notebook-book on specifying, generating, verifying, and refining multi-tenant… | — |
| beed-ml-study | Provenance audit and leakage-aware benchmark of seizure-type classifiers on the Bangalore EEG Epilepsy Dataset — IIT… | — |
| boston-house-price-prediction | Boston house price prediction with machine learning: scikit-learn gradient boosting, repeated K-fold CV, race… | open |
| Brain-Computer-Interfaces-and-Cognitive-Computing-An-Introduction | Brain–computer interfaces (BCIs) link neural activity to external devices, translating brain signals via algorithms… | — |
| building-a-virtual-cell-in-python | A cell has a normal gene-activity pattern. Knock out one gene and that pattern changes. This project trains a small… | — |
| california-house-price-prediction | California house price prediction with machine learning: scikit-learn gradient boosting, repeated K-fold and spatial… | open |
| Chronic-Kidney-Disease-Stage-Identification-in-HIV-Infected-Patients-Using-Supervised-ML | ML is transforming Chronic Kidney Disease (CKD) detection in HIV patients, catching early asymptomatic stages to… | — |
| cobratate | Cobratate, an esoteric imperative programming language. Tate-flavored syntax, ordinary semantics, and a Python… | open |
| coda-synth | CoDA-Synth: stylometric detection of LLM-rewritten dark-web documents with sparse TF-IDF + LinearSVC. Research lab,… | open |
| IIT-ROPAR-GENAI-FOUNDATIONS | Free open-source Generative AI foundations ebooks (2026) covering Claude, Grok, Gemini, Cursor, OpenAI and more —… | — |
| kbc-game | Kaun Banega Crorepati (KBC) quiz game for the terminal in Python: 16-question prize ladder, safe levels and 4… | — |
| knn-classifier | Simple K-Nearest Neighbors classifier on the classic Iris dataset using scikit-learn. Loads data, trains the model,… | — |
| linear-regression-model- | Supervised linear regression on the scikit-learn diabetes dataset. A single feature (BMI) is used to predict… | — |
| logistic-regression | This graduate capstone applies multi-class logistic regression to the Iris dataset. It includes EDA, stratified… | — |
| logistics-website-capstone | Logistics website capstone: Kargonaut Logistics, a premium freight site mixing claymorphism, glassmorphism and… | open |
| lovelight-leap | 💖 Lovelight Leap: a love-themed 2D platformer in vanilla JS canvas. 4 worlds, a boss you cheer up, original… | open |
| machine-learning-part-1 | Machine learning in Python with scikit-learn: heart disease classification and medical insurance cost regression… | — |
| machine-learning-part-2 | Computer vision with YOLO11: detection, segmentation, pose & potato leaf disease classification | — |
| machine-learning-part-3 | Road damage detection with YOLO11: potholes, cracks & manholes, in-browser demo | ▶ open |
| Meeting-of-Machines-Consumer-Agents-Seller-Agents-and-Hidden-Fulfilment-Risk-in-Online-Retail | Agentic commerce lets shopping and seller agents place orders without showing operational truth. This IS case argues… | — |
| morocco-city-aware-load-forecasting | City-aware short-term load forecasting and peak detection on high-resolution Moroccan smart-meter data (5 zones).… | — |
| NIELIT-DATA-SCIENCE | Hands-on NumPy, Pandas, and Matplotlib Jupyter notebooks for the NIELIT Data Science curriculum — practical,… | — |
| NIELIT-MACHINE-LEARNING-PROJECTS | Nineteen end-to-end machine learning projects: classification, regression, clustering, and recommendation, with… | — |
| NIELIT-PYTHON- | Open-source Python teaching notebooks: a language reference and five capstone programs, with detailed documentation.… | — |
| pan-cancer-classification | Leakage, equivalence, conformal reliability and data efficiency on the UCI TCGA Pan-Cancer RNA-Seq dataset — IIT… | — |
| Predicting-Hot-Events-in-the-Early-Period-through-Bayesian-Model-for-Social-Networks | This notebook details the implementation and evaluation of a Bayesian perspective Early stage Event Prediction… | — |
| prediction-as-power | Laboratory and working paper on prediction as power in European electricity markets. A DE-like hourly simulator… | — |
| real-estate-valuation-taiwan | Taiwan real estate valuation (Xindian, New Taipei): house price per ping prediction with scikit-learn gradient… | open |
| snakes-and-ladders | 🐍🪜 Original Snakes & Ladders browser game: hand-drawn SVG snakes and ladders, 3D dice, 2-4 humans or bots, sound,… | open |
| spam-scan | Explainable spam, phishing & Indian-scam detector for SMS/email (English, Hindi, Hinglish): TF-IDF + Logistic… | open |
| vedic-math-hybrid-kt | Hybrid Vedic Math AI: rule-based sutras (Urdhva-Tiryagbhyam, Nikhilam, Yavadunam) plus RF classifier that cleans… | — |
| wind-turbine-power-model | Single-turbine power curves + wind-farm simulation. Weibull wind stats, hourly NASA-style speed data, and… | — |
| zoo-website-capstone | Capstone project: a fictional zoo website built twice — a beginner-style simple version and a cinematic advanced… | open |