Machine learning could become a new weapon in the fight against Medicare fraud.
Machine learning can be a useful tool in detecting Medicare fraud, according to a new study that can recover anywhere from $ 19 billion to $ 65 billion lost in fraud each year.
Researchers at Florida Atlantic University’s College of Engineering and Computer Science recently published the world’s first study using Medicare Big data, machine learning, and advanced analytics to automate fraud detection. They tested six different machine learners on balanced and unbalanced data sets and eventually found that the RF100 Random Forest algorithm would be most effective in detecting potential cases of fraud. They found that unbalanced data sets are more than balanced data sets when scanning for fraud.
T
here are many implications in determining what fraud is and what is not, such as clerical error,” says Richard A. Bowder, senior author and Ph.D.
“Our goal is to allow machine learners to know all this data and flag anything suspicious. Then we can alert researchers and auditors, who should focus on 50 cases instead of 500 cases or more.”
In the study, Bowder and colleagues examined Medicare data, covering 37 million cases from 2012 to 2015, for incidents such as patient abuse, neglect, and billing for medical services. The team has reduced the data set to 3.7 million cases, which is still a challenge for human researchers charged with pinpointing Medicare fraud.
The authors used the National Provider Identifier — a government-issued ID number for health care providers to compare fraud labels with Medicare Part data, which includes provider details, payment and charge information, policy codes, all policies, and medical specifications.
When researchers compared NPI with Medicare data, they flagged fraudulent providers in a separate database.
“If we can accurately assess the physician’s uniqueness based on our statistical analyses, then we can detect exceptional physician behaviors and flag as much fraud as possible for further investigation,” said Tagi M. Khoshgofthar, Ph.D., co-author, and professor at the school.
So, if a cardiologist is wrongly labeled a neurologist, it is a sign of deception.
However, the data set remains a challenge. A small number of fraudulent providers and a large number of onboard providers have made data imbalance that can fool machine practitioners. So, using random undersampling, the researchers reduced the set to 12,000 cases, with seven class distributions ranging from severe to unbalanced.
From there, they unleashed their learners and reached their results with respect to the random forest and class distribution.
Surprisingly, the researchers found that keeping the data 90 percent simple and 10 percent fraudulent is a “sweet spot” for machine learning algorithms that work to detect Medicare frauds. They felt this proportion needed to include more fraudulent providers in order for learners to be effective.
Want to know more about Ai services then have a free visit for USM systems
