Technology & Innovation


Carlos Barge

It depends. Firstly, it’s actually really easy to detect some fraud in real-time. There are times when even a basic system using rules and no machine learning will be able to tell, because it fits all the classic patterns for fraud.

Secondly, there’s a potential distinction between ‘detect’ and ‘decision’. For example some companies provide scores in real-time. That’s a form of detection, it’s telling you that there’s effectively x% chance that this is fraud. To be able to detect fraud:

a) You must have access to a real-time stream of transactions.
b) You must build a data pipeline, and implement real-time feature engineering system that extends, enriches, and prepares the real-time stream for ML algorithms.
c) You must record real-time stream so that you can also generate features from past transactions (i.e, replaying real-time system over system logs) as your training data – if an online learning methodology is not suitable.
d) You must ensure that your real-time data pipeline generates exactly the same data as your batch pipeline, or share code between both stages (to prevent train/score set mismatch).
e) If you are detecting anomalies, you need to build a moderation system so that some experts should manually refine the anomalies against false positives (e.g., by actually contacting the customers who made the transaction etc.)
f) If you are building a classification system, you need a feedback system that requires collecting and marking past fraudulent data as positive examples in your training data set.
g) You need to form a team that tunes models, features, and moderation mechanisms to increase the accuracy continuously.

As this is a very tedious and complicated process, real time fraud prediction is possible but a difficult task. Of course, there are some types of fraud where the data is either too complex (e.g., photos of motor vehicle damages in insurance fraud), or there is no digital evidence (e.g., due to private communication of phishing victims) to be able to build an accurate ML model.

Free Pre-Assessment Request

Do you want to know how your competitors are doing business?

Tell us a little about yourself below to gain data for free

Hi What’s your name?


Hi [First Name], what is your company’s name and website?



Is your company looking for any data on the following services:



Gotcha! Do you want to monitor any specific competitor or market?

List of Competitors

  • Add competitor…



Finally, what’s your email address and your phone number?



Your Data is on the Way!

Our data scientists team is working for you by collecting data and we’ll come back to you shortly with a pre-assessment and proposal.

WYgroup BI uses the information you provide to us to contact you about our relevant content, products, and services . You can unsubscribe from communications from HubSpot at any time. For more information, check out WYgroup’s Privacy Notice.

Leave a Comment: