Machine Learning Solutions was founded to provide rapid development of custom solutions for big data
problems requiring the application of advanced analytics. Our unique approach is enabled by a
database system built from the ground up for handling big data and implementing complex analytics.
We have used this system for over 15 years to develop fraud detection and other algorithms based on
the analysis of billions of health care claims from Medicare, Medicaid and large private insurance
We use an iterative approach to big data challenges, where distinct characteristics of the data are discovered as the solution evolves – an accelerated version of “agile” development. This approach prevents any surprises at the end of the project, since preliminary test results are reviewed at each iteration to insure a good fit for with the data, the analytical approach, and the business objectives. Revealed characteristics of the particular data helps us to discover important paradigm shifts in order to increase efficiency and minimize complexity. In this way, we avoid the wasteful constraints of general- purpose applications, and accomplish the business objective in the simplest, most effective manner possible.
Our unique approach, in combination with our big-data analytical system, enables us to always find the most cost-effective solution methodology. For problems that require a certain level of performance, it will be achieved using a hardware platform costing a fraction of that possible with more traditional technologies. For a fixed hardware budget, the highest possible level of performance will be achieved. In some cases, our technology makes a solution possible that would otherwise be infeasible (see Case Study).
A major challenge in healthcare applications has always been the clustering of providers (physicians,
labs, DME suppliers, etc.) based on the mix of medical services performed and diagnoses treated. A
typical problem will involve billions of claims from hundreds of thousands of providers, who utilize tens
of thousands of codes for services and diagnoses. A hierarchical clustering problem with this many
entities and variables had previously resisted any efficient solution.
We studied the unique characteristics of this problem, and discovered an adaptation of k-nearest neighbors that solves this problem in only a few hours on relatively small computers using standard Intel processors. This solution involves three paradigm shifts from the methodologies used in traditional cluster analysis. The result has been a major advancement in fraud detection technology, identifying aberrant and/or suspicious providers in a much faster and more comprehensive manner than ever before. The clustering also creates more meaningful provider peer groups, which enhances the effectiveness of other detection algorithms that rely on peer-to-peer comparisons.
Admittedly, the clustering technique developed in this manner doesn’t have general application. However, big data problems like this are important enough to justify custom solutions. At the same time, after such a solution is developed, it is often found that it does actually fit a larger class of problems. For example, our new clustering technology can be used efficiently for any problem involving a sparsely populated matrix of continuous variables. In addition, the knowledge gained from this endeavor can be used as the basis for further adaptations for problems that don’t entirely fall within that definition. In this way, a base of knowledge and techniques accumulate, and are reused or modified in solutions to future problems. We have used a similar approach to develop custom applications in other areas, including network (fraud ring) analysis, aberrant trend detection, and several versions of multivariate predictive analysis. In the process, we have accumulated a library of methods and paradigm shifts that greatly enhance the performance of big data applications.
Machine Learning Solutions was founded in 2014 by Gene D’Angelo of Boynton Beach, Florida. Gene has
over 35 years of problem-solving experience, and over 11 years of advanced education in business,
science and engineering. It is the combination of his skills and the power of the HOPS database system
that makes possible the discovery of innovative paradigm shifts in the techniques utilized in the
application of advanced analytics in big data environments.
While working on his Ph.D. in Artificial Intelligence at the University of Central Florida in the early 1990’s, Gene took a position as an internal consultant at Blue Cross and Blue Shield of Florida (BCBSFL). In that capacity, he first encountered the problems of big data in the Medicare Fraud Branch, where they were so deluged with claims data, the simplest statistical analysis was a challenge. Gene redesigned their analytical operations and implemented a network analysis algorithm that immediately discovered many millions of dollars in fraudulent operations, including a 200-million dollar fraud ring of 165 healthcare providers in south Florida.
Following the success at BCBSFL, Gene joined HOPS International and pioneered a division with an emphasis on healthcare fraud detection. Over the next 15 years, he designed hundreds of big data algorithms for Medicare, Medicaid and large private health insurance companies. In addition to network analysis, Gene implemented many multivariate predictive models, trend analyses, pattern detection algorithms, and other methods to identify fraud, waste and abuse in healthcare.
As the demand for big data analytics grew, it became obvious to Gene there was an unmet need to find new ways of implementing advanced analytics without the constraints of the general-purpose, memory- bound algorithms used in the past for smaller problems. He realized that the database software and iterative solution methodology, that had evolved within HOPS for advanced analytics on big data problems in healthcare, was generally applicable to problems in any application domain. At that point, Gene began a campaign to include this niche in the strategic planning of his employer.
In 2012, the healthcare analytics division of HOPS was acquired by Performant Financial Corporation, where Gene assumed the position of Senior Director of Analytics Innovation. Over the next 2 ½ years, he created analytic products that generated millions of dollars of new profits for Performant. Most of these products were multivariate predictive models designed to allocate auditing and problem resolution efforts to areas with the highest potential profitability. One of the most powerful systems Gene created during this period was a big data approach to Classification and Regression Tree (CART) multivariate analysis.
It soon became apparent that Performant was not prepared to take advantage of the potential of HOPS and the iterative development approach to big data problems. For this reason, Gene founded Machine Learning Solutions in early 2014 as a part-time endeavor. In November of 2014, he left Performant to concentrate full-time on the challenges of big data analytics.