Accenture: Supply Chain Resilience

Inspiration

Increasing globalization poses challenges at a new scale. We aim to leverage recent advances in ML to cope with these developments.

What it does

It allows companies to predict the likelihood of an order being delayed based on inputs regarding the properties of the order.

How we built it

After extensive data exploration, we identified and engineered key features, which we then used on multiple approaches such as Nearest Neighbors, XGBoost, AdaBoost, and a neural network. We combined these approaches using ensemble modeling methods. In production, the frontend calls this model and returns the probability of a delayed order.

Challenges we ran into

Class imbalances and narrowing down the problem scope.

Accomplishments that we're proud of

Reaching a good accuracy and building a complete MVP.

What we learned

This project allowed us to get more hands-on practice in combining theoretical ML knowledge with forming a business case and thinking from a real-world perspective.

What's next for Accenture: Supply Chain Resilience

We would continue by implementing a report download feature that give the user a granular way of accessing the already existing insights on details that are generated when providing inputs about the order.

Built With

• pandas
• pydeck
• python
• scikit-learn
• streamlit