02 — PROBLEM
Problem
Build a modular classification engine that can compare several candidate algorithms while keeping preprocessing and inference paths consistent.
PROJECT / 03
Credit Risk Classification And Deployment Engine
A credit risk classification engine built around applicant financial and demographic data. Five algorithms were benchmarked before the selected models were serialized and exposed for production inference.
SYSTEM PATH
01DATA
02MODEL
03EVAL
04API
05CONTAINER
06CI/CD
07CLOUD
5algorithms benchmarkedclassification models
83.12%validation accuracyGaussian Naive Bayes + SVC
01 — SYSTEM ARCHITECTURE
How the pieces connect.
The case study keeps architecture data-driven too. Each project can describe a different system path without changing the renderer.
01
Applicant financial + demographic data
02
Categorical encoding + feature scaling
03
Train/test split
04
5-model benchmark
05
Serialized selected model
06
FastAPI + Pydantic → Docker → AWS EC2
03 — APPROACH
Solution
Use a preprocessing pipeline for categorical encoding, feature scaling, and train/test splitting across income, credit-history, and property-area features, then benchmark five algorithms.
04 — RESULTS
Results
Gaussian Naive Bayes and SVC both achieved 83.12% validation accuracy in the benchmark.
05 — IMPLEMENTATION
Built for the path to deployment.
01
Compared Random Forest, Gaussian Naive Bayes, Decision Tree, KNN, and SVC.
02
Serialized the selected model for production inference.
03
Added Pydantic request validation at the FastAPI boundary.
04
Containerized the service with Docker and hosted it on AWS EC2.
06 — STACK & DEPLOYMENT
FastAPI + Pydantic + Docker + AWS EC2