Machine Learning · Blockchain · KYC/AML · 5 notebooks + bonus
Blockchain Fraud Detection
Approach fraud detection as a real data problem: understand transactions, describe wallets, use graph relationships and evaluate a model despite the scarcity of illicit cases.
PythonPandasNetworkXXGBoost
Free · No account·Advanced·~3h
What you'll build
- ✓A complete pipeline turning blockchain transactions into a risk score for each wallet
- ✓Behavioral indicators that summarize address activity and isolate bots
- ✓A graph view of transactions to measure proximity to known illicit wallets
- ✓A rigorous comparison between a linear baseline and XGBoost under severe class imbalance
Lab notebooks
01
Understand transaction data
Explore amounts, timing, available labels and the first structures of the transfer network.
02
Describe wallet behavior
Aggregate transactions, compare time windows and identify automated activity.
03
Reason over the graph
Measure distance to known illicit wallets and study how bot filtering affects propagation.
04
Model an imbalanced problem
Use a temporal split, establish a baseline and compare models with appropriate metrics.
05
Consolidate and score
Run the full pipeline end to end and score wallets that have not yet been labeled.
06
Consider real-time scaling
Explore Kafka, Spark Streaming and incremental propagation through an architecture and simulations.
Skills covered
- · Transaction data analysis
- · Entity-level feature engineering
- · Graph analysis and traversal
- · Class imbalance management
- · Baseline and advanced-model comparison
- · Precision-recall and ROC curve evaluation
Prerequisites
- · Intermediate Python (Pandas, loops, functions)
- · Basic supervised machine learning knowledge
- · No installation : online lab