Simplifying Game Fraud Detection
Designed a machine learning approach to identify collusion in game activity.
- Fraud Detection
- Machine Learning
- Game Systems
- System Design
Nitin Goel
Senior Software Engineer at Atlassian
From their time as
Software Engineer - II
Moonfrog Labs • 2019 - 2020
Overview
Nitin designed a fraud detection approach for a game serving millions of daily users. He replaced deeply nested rules with a machine learning decision layer for player-pair collusion.
The Story
Nitin worked on fraud and cheating detection for a game serving millions of daily users. Collusion patterns could be described with a growing set of conditional rules, but that approach would become difficult for new engineers to understand and maintain.
He chose a machine learning approach instead. Game actions were supplied to a model that assessed whether a pair of users appeared to be colluding.
The design moved the complexity out of large nested conditional statements and into a decision layer built for the classification problem. It gave the team a clearer way to evolve fraud detection as behavior changed.
Ownership Snapshot
Broad role
Full-stack engineer on a large-scale game.
Goal
Detect cheating and collusion without complex rules.
Direct ownership
Designed the fraud detection approach.
Team execution
Built within the game engineering team.