Skunkworks All-Hands ML Detectives Activity

Author: Benjamin Afflerbach
Date Uploaded: 2025-05-10
Content Type(s): Workshop Activity
Content Length: 1-2 hrs
Content Audience: undergraduate
Content Topics: Machine Learning, Overfitting, Dataset Drift, Data Labeling, and Data Leakage

---

Overview

This is a 1-2 hour group activity developed for the Spring 2025 Informatics Skunkworks All-Hands meeting. It is meant to be completed in small groups as a competition in which each group is a team of “Machine Learning Detectives” trying to solve a series of cases related to why a machine learning model is performing poorly. There are 6 evidence packs that teams are instructed to hunt through round by round during the activity. When a team thinking they’ve found the problem they fill out the Case Report document then give a 1-2 minute explanation to the group about what they found.

Learning Objectives

• Students interpret common Machine Learning Assessment plots and error metrics
• Students compare inputs and outputs to identify conflicting information describing a machine learning model’s performance
• Students present results and justify their conclusions

Additional Information

The activity is currently setup with a canned set of results that aren’t directly linked to a particular real world dataset. A more compelling setup with more dev time might use a real world dataset and show these effects actually happening with real interactive code.

Links

Shared Google Drive with evidence packs and case report document template

---

Return to Collection

Return Home