Overview
Believe it or not, 1/3 of all food produced in America goes to waste. Restaurants and grocery stores alike often throw away expired food that didn’t get used because of bad planning or overly optimistic outlook.
Using machine learning algorithms, we created an app that makes predictions about the food consumption of upcoming weeks. The app learns individual stores’ order patterns, and fuse them with macro-level live data such as weather, market supply, and local events.
WasteNot! is a Global Finalist of NASA Space Apps Challenge 2017. Visit our project page here. And watch our presentation here.
MY ROLE
Being the sole product designer, I lead our team to analyze the current food ecosystem, and to brainstorm how to use machine learning to maximize our impact on reducing food waste. Once we designed a product model, I worked with the back-end engineer on developing feasible product features, and with the front-end engineer on designing a user-friendly interface.
CHALLENGES
The core of our product is to use machine learning algorithms to learn patterns by comparing local-level food consumption data and macro-level environmental/social data. Because we didn't have any food supplier to work with at the hackathon, we used simulated restaurant data sets. Going forward, we plan to work with real businesses to do test runs of our system.
Analysis and Feature Design
We started the design process by mapping out the life cycle of food, and the stakeholders inside the ecosystem. A lot of the food waste comes from expiration, so we looked into the relationship between demand and supply -- why were there extra food stocked up for the consumption that never happened? How do we find patterns in data sets that might predict future consumptions?
Data fusion and machine learning were the answer.
By integrating individual restaurant data with macro-level datasets like weather, market supplies, and cultural events, the machine learning algorithm can predict consumer behaviors based on the local consumption patterns. Through time, the software learns the patterns through a feedback loop, and make predictions with higher confidence.
