API_Based_Prediction

A project where a scikit-learn model is wrapped in a three-tier service architecture: Python frontend → Go (Gin) gateway → FastAPI with the model. The whole stack comes up with a single docker compose command.

The point of the project isn't the model itself — it's the end-to-end exercise: splitting responsibilities across services, getting them to talk over HTTP, and packaging everything so it runs reproducibly anywhere.

Why three services for one model?

To get hands-on with a pattern that shows up a lot in production: a thin client, a fast gateway in front, and a heavier Python ML backend behind it. Two practical wins from this layout:

• The gateway can be scaled or have caching/rate limiting added independently of the heavier Python process.
• The ML service is isolated, so the model can be retrained and swapped without touching the rest of the stack.

The model

Task. Binary classification - given a person's profile, predict whether their credit score qualifies as High (label 1) or not (Average/Low collapsed to -1). Dataset: Credit Score Classification Dataset (Kaggle), kept locally in data/.

Algorithm. LogisticRegression from scikit-learn (max_iter=1000). A linear model is a natural choice here: the dataset is small, the feature set is low-dimensional, and a logistic regression gives me interpretable coefficients and a calibrated predict_proba — which is exactly what I need to report ROC AUC.

Features and preprocessing.

• Education and Number of Children dropped - weak/irrelevant signal for this target.
• Categorical features encoded as {-1, +1}: Gender, Marital Status, Home Ownership. With only two categories each, this is equivalent to one-hot but keeps the feature matrix narrower.
• Income → log1p to compress the right tail, then clipped at the 99th percentile to limit the influence of outliers, then standardized with StandardScaler.
• Age standardized with StandardScaler.

Running it

Option 1: Docker Compose (recommended)

docker compose up --build

Brings up all three services at once.

Option 2: Locally, service by service

pip install -r requirements.txt

# ML service (FastAPI)
uvicorn python_api.app:app --host 127.0.0.1 --port 8000 --reload

# Gateway (Go + Gin)
go run go_api/main.go

# Frontend
python web/main.py

Stack

Python · FastAPI · scikit-learn · Go · Gin · Docker · Docker Compose · Jupyter