Datamatics aimed at developing a sophisticated rules engine that would not just be capable of detecting fraudulent claims in real time through algorithmic interventions, but also capable of improving itself leveraging machine learning with increased exposure to data.
Datamatics studied the raw data set vis-à-vis both the claim values as well as the invoice lines. The team segregated the attributes in terms of Age, Claim Invoice Gross GBP, and Claim Invoice Net GBP. They also divided the attributes as Gender, Country Code, Diagnostic Code, Claim Status, Policy Type, etc., for the actual values.
Datamatics implemented an ensemble of machine learning techniques coupled with anomaly detection to analyse 3.3 million claims to identify potentially fraudulent claims within the formed clusters. Complex technological interventions utilized to devise a user-friendly interface capable of detecting fraudulent behavior in real time.
Granular line item level data comprising of several attributes in all possible scales (nominal, ordinal, interval and ratio) was utilized and challenges related to non-availability of markers were addressed with ingenuity.