Manufacturing Claims Anomaly Detection (Batch + Drift)

This project demonstrates a production-style batch ML workflow for detecting anomalies in manufacturing-related data such as sales, claims, and equipment signals.
It includes: -- Feature engineering -- Anomaly detection model training (Isolation Forest) -- Batch inference (scheduled scoring) -- Simple drift detection using statistical distribution checks

Why this project

In manufacturing environments (e.g., motorboats, motorhomes, towables), abnormal patterns in claims or equipment signals can indicate quality issues or operational risks.
This repo simulates a realistic workflow used in enterprise settings where batch inference is preferred for cost and stability.

Data

The repo generates a synthetic dataset that mimics: -- Sales units -- Warranty claims -- Equipment temperature and line speed -- Product, plant, and region attributes
An incident window is injected to simulate a real anomaly spike.

Quickstart

python -m venv .venv
source .venv/bin/activate  # (Windows: .venv\Scripts\activate)
pip install -r requirements.txt

python src/generate_data.py
python src/train.py
python src/infer.py
python src/drift_check.py