PaySim — Fraud Analytics and Transaction Monitoring
Context
PaySim is a synthetic mobile-money transaction dataset, built to model realistic financial behaviour without exposing real customer data. It’s a widely used benchmark for practicing fraud analytics on transaction-level data.
Objective
Identify transactions consistent with common fraud typologies — structuring, unusual velocity, behavioural anomalies and unfamiliar counterparty relationships — using a complete analytics workflow rather than a single detection rule.
Methodology
Each typology is treated as its own detection problem rather than folded into one generic fraud score. Structuring and velocity are approached as SQL-level, rule-based logic; behavioural anomalies and counterparty risk rely more on feature engineering and exploratory analysis in Python. The two approaches are brought together in Power BI, where flagged activity can be reviewed and cross-checked by type, time and amount.
Workflow
- Data preparation — cleaning and validating transaction records, checking for missing values, duplicates and inconsistent types.
- SQL — window functions and rule-based flags, e.g. isolating rapid sequences of transactions or unusual counterparty patterns.
- Python / Pandas — feature engineering and exploratory analysis on the prepared data.
- Power BI / DAX — interactive dashboards summarising flagged activity by type, time and amount.
Key findings
The workflow isolates transactions matching each of the four target typologies and separates them from legitimate activity using explainable, rule-based and statistical criteria. Because the logic is built in SQL and surfaced in Power BI, each flag can be traced back to the condition that triggered it — the kind of reasoning an analyst needs to be able to justify an escalation decision.
Tools
SQL · Python · Pandas · Power BI · DAX
Limitations
PaySim is a synthetic dataset with labelled, synthetic fraud cases. This project demonstrates an analytical methodology, not a production fraud detection system, and is not based on real customer transactions.