AML: Reducing False Positive to improve regulatory reporting

Challenges.

• Increasing AML activities and complex customer behavioral events results into decreasing detection rates
• Regulatory compliance needed to be adhered to and this involves  a lot of manual efforts.
• Need to monitor and check the compliance databases from OFAC and other sources to catch the anomalous behavior 
• Transaction rules are not sufficient as the behavior keeps on and generate a lot of false positive

Solution.

• Algorithm Used: Soft clustering behavioral misalignment score
• Aggregate banking transactions and build customer clusters
• Suspicious customers are: not in any cluster, misaligned within the clusters and deviate from the cluster prototype
• Use ML classification prediction model using historical False Positive alerts and related information as input.
• Check against compliance databases using techniques like fuzzy and regex matching
• Apply on any new input to get a classification score

The Impact.

• Significant fall in false positives
• Efficient fraud management and Improvement in fraud detection rates 
• Effective workload management