Separator

Next Level Analytics

Separator
Christopher Westphal, Chief Analytics Officer, DataWalk California based DataWalk is a software platform to connect numerous large data sets, both external and internal, into a single repository for fast visual analysis.

What’s the cure for cancer? How is the stock market going to perform? Who committed the murder? The answers to these questions are already contained in the data; but how do we find them? Many people are turning to Artificial Intelligence (AI) and Machine Learning (ML) for these answers. AI/ML is transforming our society. These powerful technologies are incorporated into all aspects of our existence including video games, social media, and smart homes. They also cover delivering improved medical advice, better healthcare services, and advanced fraud detection. Without a doubt, many of these advances make our lives better, more comfortable, and arguably, more secure, but there’s more to be done.

AI/ML are well suited to solve complex, well-defined, analytical problems that specifically differentiate positive, negative, or general categorical outcomes. With enough examples and training cycles these systems can often exceed human levels of expertise, which is invaluable, especially in real-time environments like self-driving cars, high-volume claims processing, or repetitive activities such as quality control in manufacturing environments. To many, AL/ML offers a panacea to solve the world’s problem and there are certainly excellent use-cases for the technology. But as Abraham Maslow once stated (1966), ‘I suppose it is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail.’ Perhaps more apropos would be the cliché ‘there's more than one way to skin a cat.’ In fact, solving today’s complex problems require a multi-faceted approach of various techniques and technologies.

There are well-documented limitations related to the common use of AI/ML. In general, they are intrinsically narrow and
limited to their designated tasking and don’t easily adapt to new problem areas without a substantial reinvestment to realign them. They require large training sets pre-tagged with the appropriate classification values so the outcome is already determined. Their models can become biased; and, they are not well suited to explain their decision-making. Yet, AL/ML still delivers some great value, but they don't necessarily represent the entire solution. I’ve personally worked on many analytical engagements for government agencies and focused on targeting criminal behavior such as terrorists, money launderers, drug dealers, or human traffickers. Analytics for intelligence, fraud, and criminal detection deal with very complicated and dynamic domains and massive amounts of data without well-defined parameters, quantitative examples, or any type of routine or common definitions. Unfortunately, there is no ‘Easy Button’ to automatically discover patterns of interests – especially when they’ve never been seen before.

A good example of this is represented in our continued losses within our monolithic healthcare system. In 2015, the US Government’s Medicare Fee-for-Service (Parts A and B) official estimates for improper payment rate were 12.1 percent, representing $43.3 billion. An improper payment is any payment that should not have been made or was made in an incorrect amount including, duplicate payments, payments to ineligible recipients, and payments with insufficient or no documentation. It also includes various types of fraud and misrepresentation. There are many examples of Medicare fraud schemes including: addresses listed for doctor’s offices or clinics that are really mailbox rental stores or vacant lots; paying kickbacks to patients including gift card, trips, or drugs; using patient-recruiters, called ‘cappers,’ to illegally obtain patients (or their Medicare IDs) to prescribe unnecessary surgical procedures or devices such as motorized wheelchairs; or those providers involved in the unlawful distribution of opioids and other prescription narcotics. There are literally thousands of different types of Medicare scams occurring on a regular basis.

To many, AL/ML offers a panacea to solve the world’s problem and there are certainly excellent use-cases for the technology

To effectively handle the range of different frauds or improper payments requires a combination of approaches including AL/ML, statistics, and entity-based analytics. There is no ‘one-size-fits-all’ algorithm that identifies all types or situations. In fact, the Centers for Medicare & Medicaid Services (CMS) have created a Fraud Prevention System (FPS) defined as using state-of-the-art predictive analytics technology. Since 2011, CMS has spent $182M in implementation costs, with over $1.5 billion identified (through 2015) as improper and potentially fraudulent payments.

2018 will be a fascinating year with tremendous technology advances in data analytics. Bigger and faster processing systems will continue; new types of hybrid computation models will emerge; and more refined analytics will become commonplace. It’s an exciting time to be part of this market place.