ML · FastAPI · Docker · Google Cloud Run

Breast Cancer
Classifier.

Logistic Regression and Random Forest models trained on the Wisconsin Breast Cancer dataset, served via a FastAPI REST endpoint, containerized with Docker, and deployed to Google Cloud Run — a full production ML pipeline.

View on GitHub See Results

Production Pipeline Cloud Run

Wisconsin Dataset 569 samples · 30 features · binary labels

Train Models Logistic Regression + Random Forest · scikit-learn

FastAPI Service REST endpoint · /predict · JSON in, prediction out

Docker Container Portable · reproducible · production-ready image

Google Cloud Run Serverless · auto-scaling · public HTTPS endpoint

Results

Model performance.

97.4%

Logistic Regression accuracy

// Wisconsin test set

96.5%

Random Forest accuracy

// Wisconsin test set

Input features

// cell nucleus measurements

569

Training samples

// malignant + benign

Logistic Regression

Simple, interpretable, fast

A linear model that learns decision boundaries in feature space. Strong baseline for binary medical classification — weights are directly interpretable as feature importance.

97.4% accuracy on Wisconsin test set
Fast inference — suitable for real-time API responses
Coefficients give direct clinical feature insight
Minimal risk of overfitting on small datasets

Random Forest

Ensemble, robust, feature-aware

An ensemble of decision trees that votes on the final prediction. Captures non-linear feature interactions and provides built-in feature importance ranking.

96.5% accuracy — strong generalisation from ensembling
Feature importance ranking across all 30 features
Robust to outliers and noisy measurements
No feature scaling required

Architecture

Full pipeline design.

Wisconsin Breast Cancer Dataset 569 samples · 30 numeric features · labels: Malignant / Benign ────────────────────────────────────────────────────────────── Preprocessing ├── Train / test split (80 / 20) ├── StandardScaler → zero mean, unit variance └── Applied to both LR and RF pipelines Model 1 — Logistic Regression scikit-learn ├── Solver: lbfgs Max iter: 10,000 ├── Regularization: L2 (C = 1.0) └── Output: P(malignant) + binary label Model 2 — Random Forest scikit-learn ├── n_estimators: 100 max_features: sqrt ├── Bootstrap sampling: True └── Output: majority vote + class probabilities ────────────────────────────────────────────────────────────── FastAPI Service ├── POST /predict │ body: { "features": [f1, f2, ... f30], "model": "lr" | "rf" } │ resp: { "prediction": "malignant" | "benign", "confidence": 0.97 } ├── GET /health └── GET /docs (auto Swagger UI) Docker ├── Base image: python:3.11-slim ├── Copies model artefacts + app └── EXPOSE 8080 → uvicorn main:app Google Cloud Run ├── Serverless container runtime ├── Auto-scales to zero when idle └── Public HTTPS endpoint → zero infrastructure management

Deployment

From laptop to cloud.

🧪

Training

scikit-learn

Models trained locally, serialized with joblib. Both LR and RF packaged as model artefacts alongside the API.

joblibpickle

⚡

API

FastAPI

Async REST service with automatic Swagger UI. Input validation via Pydantic. Returns prediction and confidence score.

PydanticUvicornSwagger

🐳

Container

Docker

Reproducible build. All dependencies, model artefacts, and the app bundled into a single portable image.

python:3.11-slimEXPOSE 8080

☁️

Cloud

Google Cloud Run

Serverless container hosting. Scales to zero when idle, auto-scales under load. Public HTTPS endpoint, no server management.

ServerlessAuto-scalingHTTPS

~ docker build -t breast-cancer-classifier .

Successfully built a3f92c1d8e04

~ docker run -p 8080:8080 breast-cancer-classifier

INFO: Uvicorn running on http://0.0.0.0:8080

~ gcloud run deploy --image breast-cancer-classifier --platform managed

Deploying container to Cloud Run service in region us-central1...

Service URL: https://breast-cancer-classifier-xxxx.run.app ✓

~ curl -X POST /predict -d '{"features":[...], "model":"lr"}'

{

"prediction": "benign",

"confidence": 0.974,

"model": "logistic_regression"

}

Stack

Built with.

Category	Tool	Purpose
Dataset	Wisconsin Breast Cancer UCI / sklearn	569 samples, 30 features, binary classification
Models	scikit-learn LR + RF	Logistic Regression and Random Forest classifiers
API	FastAPI + Pydantic	REST endpoint with automatic validation and Swagger docs
Server	Uvicorn	ASGI server for FastAPI
Container	Docker python:3.11-slim	Reproducible, portable deployment image
Cloud	Google Cloud Run serverless	Zero-ops container hosting with auto-scaling
Serialization	joblib	Save and load trained model artefacts

Breast CancerClassifier.