Trulioo — Solution Sheet
100% True Accept
and 100% True
Reject Rates Are
Possible
Machine learning (ML) revolutionized
the way we combat fraud. With its
ability to detect suspicious activity
faster and more accurately than
traditional manual checks, ML is a
valuable asset in the fight against fraud.
Using document verification ML models, businesses
can train their systems to differentiate between real
and fake identities. Those models are evaluated using
a confusion matrix, which measures the acceptance
and rejection rates of legitimate and fraudulent
documents.
High true accept rates (TAR) and true reject rates
(TRR) are crucial for businesses to successfully
onboard legitimate users and keep fraudsters at
bay. However, businesses should be cautious when
vendors tout a perfect confusion matrix.
In reality, there is always a tradeoff between TAR and
TRR. Setting the model threshold to accept 100% of
legitimate users will inevitably result in more fraudsters
slipping through the cracks. A perfect confusion matrix
could indicate the models were not tested on diverse,
real-world data.
It’s important to remember that ML is probabilistic, not
deterministic. False positives and false negatives are
inevitable, and verification results can be influenced by
factors such as document type and image quality.
As ML evolves and improves, businesses can make
informed decisions about how to optimize their
verification processes and strike the right balance
between accepting legitimate users and keeping
fraudsters out.
03
Myth