South African cryptocurrency platform OVEX has put a new fraud-detection system into operation, using machine learning developed with DataProphet to examine how accounts behave over time.
The companies began working together in April. Their approach follows sequences of activity, including logins, deposits, trades and withdrawals, rather than relying only on a single transaction or a fixed list of warning signs. Separate models examine patterns associated with legitimate customers and fraudulent activity, helping investigators decide which accounts need closer attention.
In testing, the system identified 98% of fraudsters, compared with 75% under the previous rules-based approach, the companies said. That is a test result rather than a guarantee of performance against every future attack. OVEX also expects fewer false alarms, which would reduce the workload created when genuine customers are incorrectly flagged.
The rollout includes tools for investigators to see when an account’s behaviour changed and for teams to monitor and retrain models. For a trading platform, the practical challenge is maintaining those controls as customer behaviour and fraud techniques evolve.