Predicting fraud leveraging data science for a large US insurer

Problem Statement
An American supplemental insurance major need to deal with fraudulent cases in ever-increasing claims transactions. The insurer was largely dealing the fraudulent activities with a rule-based approach of manual processing and investigation (driven by expert judgement of agents, investigators and auditors). They wanted to move to a more scientific approach leading to strategizing the next generation Fraud Analytics System.

Solution Overview

  • Machine learning-based predictive model devised
  • Implementing feature engineering techniques
  • Segmentation to detect patterns
  • Supervised learning on specific clusters to measure its strengths and further strengthen the model
  • Tool stack used: Python, Splunk

Outcomes

  • 500 basis points increase in the accuracy of Fraud detection
  • Moving from rule-based fraud identification to automated way using Machine Learning 
  • Ability to significantly scale up the number of claims processed

    Image