Authors: Charlie Berger and Dohoon Kim
Automotive insurance fraud involves someone deceiving an insurance company about a claim involving their personal or commercial motor vehicle. It can involve giving out misleading information or providing false documentation to support the claim.
Nearly one of 10 Americans would commit insurance fraud if they knew they could get away with it. Nearly one of four Americans say it’s ok to defraud insurers. About one in 10 people agree it’s ok to submit claims for items that aren’t lost or damaged, or for personal injuries that didn’t occur. Two of five people are “not very likely” or “not likely at all” to report someone who ripped off an insurer. Accenture Ltd.(2003) Nearly three of 10 Americans (29 percent) wouldn't report insurance scams committed by someone they know. Progressive Insurance (2001)
This blog highlights the use of two Oracle Machine Learning notebooks working in conjunction with automobile insurance claims investigators in a two-step process. First, we use Oracle Machine Learning to “flag” for the investigator anomalous insurance claims using an unsupervised learning algorithm (1-Class Support Vector Machine). We help the claims investigators focus on the most suspicious claims using their expertise and knowledge using an Oracle APEX application.
After their investigation of the most suspicious claims, we ask them to enter their decisions (Fraudfound? Yes or No) creating "labeled" data for building supervised learning classification models on the target attribute (FraudFound) using Oracle Machine Learning. Based on this two-step process of unsupervised learning and supervised learning combined with human expertise, we can build a data and ML-driven methodology to detect costly fraudulent auto claims.
Below are highlights from two Oracle Machine Learning notebooks, Oracle APEX and Oracle Analytics Cloud.
Watch "A Two-Step Process for Detecting Fraud using Oracle Machine Learning" presentation and OML + APEX and Oracle Analytics Cloud integrated demos on YouTube.
Download the presentation.
We use Oracle Machine Learning's 1-Class Support Vector Machine to train on "normal" records and apply the 1-Class SVM model to our insurance claims to flag those claims that are most dissimilar from the training population. Note: If you have known fraudulent cases, you should remove them from the training dataset.
Based on our "unlabeled data" we've built and applied Oracle Machine Learning models that identify the most suspicious claims and have provided OML's reasons why (Prediction_Details). Now we want to support the claims investigators so they can focus on the most suspicious claims. Using their knowledge and years of expertise, after their investigation, they can use a simple Oracle APEX application to enter their determinations as to whether the claim is fraudulent or not (Fraudfound - Yes or No). We will use this new knowledge to build better predictive ML models.
Using this curated "labeled" data we build, evaluate and apply supervised learning models on the target attribute (FraudFound - Yes/No) using Oracle Machine Learning's classification models. The following OML screenshots show highlights of this second step of our two-step ML methodology.
Now, we can leave our data, ML models, predictions and insights inside Oracle Autonomous Database and review them using Oracle APEX.
Finally, we can interactively explore our most likely automobile insurance claims using Oracle Analytics Cloud ane its wide range of interactive tables, charts, dashboards and filters.
Based on this two-step process of unsupervised learning and supervised learning combined with human expertise, we can build a data and ML-driven methodology to detect costly fraudulent auto claims.
"Fraud Journey: Human Expert and Machine Learning Working Together to Detect Auto Claims Fraud" at Analytics and Data Oracle User Community TechCasts, Oct. 13, 2020 by Dr. Abi Haigh and Charlie Berger