Fraud Detective Challenge | Anomaly Detection Activity

🎯 Activity Goal

Your team is a fraud investigation unit. You'll analyze transaction data to identify anomalies, design your own fraud patterns, and try to stump other teams — all while learning supervised and unsupervised anomaly detection thinking.

🗂️ Choose Your Scenario

Select the industry scenario your team wants to investigate:

🏦

Banking

Credit card transactions with potential fraud

🏥

Healthcare

Insurance claims with suspicious billing

🎓

Campus

Student dining & access card swipes

🛒

E-Commerce

Online orders with return/refund abuse

📖 How It Works

Round 1 — Detect (10 min): Examine 30 transactions. Flag the ones you think are anomalies. Some have labels (supervised clues), others don't — you'll need to spot patterns on your own (unsupervised thinking).

Round 2 — Plant (5 min): Your team designs 3 fraudulent transactions to hide in a clean dataset. Make them clever enough to fool another team!

Round 3 — Swap & Hunt (10 min): Exchange planted datasets with another team. Find their hidden fraud. The harder your fakes are to detect, the more points you earn.

⏱️ 10:00

🔎 Round 1 — Detect the Anomalies

Review the transactions below. Click "Flag" on any row you believe is anomalous. Look for unusual amounts, odd timing, location mismatches, or behavioral patterns that don't fit.

💡 Hint: The first 5 transactions have a "Risk Label" column (supervised). The rest don't — you're on your own (unsupervised). Think about what features or rules you'd use.

Team Name:

#	Date/Time	Customer	Amount	Category	Location	Risk Label	Action

⏱️ 5:00

🎭 Round 2 — Plant Your Fraud

Now you are the fraudster! Design 3 sneaky anomalous transactions that will blend into a normal dataset. The other team will try to catch them. Make them realistic but subtly off.

💡 Strategy Tips: Avoid obviously huge amounts. Think about subtle mismatches — a purchase at 3 AM in a different city, a category that doesn't match the customer profile, or a velocity anomaly (many transactions in a short window).

Planted Transaction #1

Date/Time

Customer

Amount ($)

Planted Transaction #2

Date/Time

Customer

Amount ($)

Planted Transaction #3

Date/Time

Customer

Amount ($)

🔄 Round 3 — Swap & Hunt

You've received another team's dataset with their 3 planted anomalies hidden among normal transactions. Work together to find all 3!

💡 Discussion Questions as You Hunt:
• What unsupervised technique would help here? (Clustering? Isolation Forest? Z-score?)
• If you had labels, what supervised model would you choose? (Logistic regression? Decision tree?)
• What features would you engineer to catch these anomalies?

Record Your Findings

For each suspected anomaly, write which transaction and why:

Suspected Anomaly #1:

Suspected Anomaly #2:

Suspected Anomaly #3:

🧠 Team Reflection (Discuss Together)

After finding the planted frauds, discuss:

Supervised vs. Unsupervised: Which approach would work better for this dataset, and why?
Feature Engineering: What new features (time-of-day bucket, amount z-score, velocity count) would improve detection?
Real-World Application: Where in your career or industry could anomaly detection create the most value?
False Positives vs. False Negatives: What's worse — flagging a good transaction or missing a fraud? How does the business context change your answer?

🏆 Scoring & Leaderboard

0

Correct Flags (R1)

0

False Alarms (R1)

0

Missed Frauds (R1)

0

Total Points

📊 Scoring Rubric

Action	Points	Notes
Correctly flag an anomaly (R1)	+10	True Positive
False alarm — flag a normal tx (R1)	-5	False Positive penalty
Miss a real anomaly (R1)	-3	False Negative
Other team fails to find your plant (R3)	+8	Per undetected plant
Other team catches your plant (R3)	+2	Good sportsmanship
Correct explanation of technique	+5	Bonus for stating which method

💼 Career Connection Rubric (Bonus Points)

Each team gets up to 10 bonus points for connecting anomaly detection to a real-world career scenario:

+3: Named a specific industry use case
+3: Identified which approach (supervised/unsupervised) fits and why
+4: Proposed a realistic feature engineering strategy

🔍 Fraud Detective Challenge

🎯 Activity Goal

🗂️ Choose Your Scenario

📖 How It Works

🔎 Round 1 — Detect the Anomalies

🎭 Round 2 — Plant Your Fraud

Planted Transaction #1

Planted Transaction #2

Planted Transaction #3

🔄 Round 3 — Swap & Hunt

Record Your Findings

🧠 Team Reflection (Discuss Together)

🏆 Scoring & Leaderboard

📊 Scoring Rubric

💼 Career Connection Rubric (Bonus Points)