Project Back

Azure AutoML vs. Custom Model Development

Completed or published: 2024-01-01

Compared Azure AutoML with a custom Jupyter-based modeling workflow for shipping-cost prediction, focusing on tradeoffs in speed, control, and model quality.

Project 2024
Azure AutoML vs. Custom Model Development
Azure AutoML vs. Custom Model Development Project overview
Azure AutoML vs. Custom Model Development
Azure AutoML vs. Custom Model Development
Azure AutoML vs. Custom Model Development

Azure AutoML vs. Custom Model Development

This project compared two ways to solve a regression problem: automated machine learning in Azure AutoML and a custom Jupyter Notebook workflow. The use case was shipping-cost prediction for art and collectible objects.

Data-science question

AutoML can accelerate experimentation, but it does not remove the need for careful problem framing. The comparison asked when a fast automated baseline is enough, and when custom feature engineering, validation design, and interpretability are worth the extra work.

Modeling context

The dataset included package dimensions, weight, transport mode, fragile-item indicators, customer and delivery locations, scheduled dates, delivery dates, and shipping cost. The target was continuous, making the project a practical regression study.

What it demonstrates

  • Building a benchmark model quickly with AutoML
  • Designing a custom notebook workflow for deeper control
  • Comparing speed, transparency, and model quality
  • Thinking about model development as an analytics decision, not just a tool choice