Credit card fraud is a major concern for financial institutions. The challenge in fraud detection is the extreme data imbalance: in this dataset of European cardholders, out of 284,807 transactions, only 492 were fraudulent (accounting for a mere 0.17%).
The goal of this project is to build an interactive Power BI dashboard that uncovers hidden patterns of fraudulent activities, answering three core questions:
Dataset: Credit Card Fraud Detection
Access the Power BI file (.pbix) and DAX implementation on GitHub.
1. Data Processing & DAX Modeling (Power BI) To extract meaningful insights from raw data, several transformations and DAX measures were applied directly within the BI environment:
Time Series Transformation: Converted raw elapsed seconds into a 24-hour cyclical format (MOD 24) to identify daily seasonality and peak hours of fraudulent activities.
Hour = MOD(INT(creditcard[Time] / 3600), 24)
Amount Binning Strategy: Instead of using raw transaction amounts, categorizations were created for specific bins (e.g., Micro <$50, Small $50-$200, Extreme >$1000) using a DAX SWITCH statement. This was crucial to isolate "card-testing" behaviors from "high-impact" thefts.
Amount Category = SWITCH(
TRUE(),
creditcard[Amount]=0,"Zero",
creditcard[Amount]<=50,"Micro(<50)",
creditcard[Amount]<=200,"Small(50-200)",
creditcard[Amount]<=500,"Medium(200-500)",
creditcard[Amount]<=1000,"High(500-1000)",
creditcard[Amount]>1000,"Extreme(>1000)",
"Other"
)
Cross-Filtering Optimization: Designed a "Corporate Dark" UI with active cross-filtering, allowing stakeholders to slice data seamlessly by Day and view isolated fraud behaviors without being overwhelmed by normal transaction noise.
💡 Key Business Insights Based on the dashboard visualizations, several critical fraud patterns were identified: