Our Beyond Buzzwords series focuses on demystifying the jargon surrounding healthcare analytics. Our most recent blog, Applications of Artificial Intelligence in Healthcare, tackles the nebulous world of artificial intelligence, machine learning and natural language processing in combatting fraud, waste and abuse. Here, we’re peeling back the layer on machine learning to explore the concept of anomaly detection and its role in facilitating payment accuracy.
Anomaly Detection: An Unsupervised Machine Learning Technique
Before diving into the specifics of anomaly detection, it is worth briefly exploring the distinction between supervised and unsupervised machine learning.
Supervised machine learning algorithms draw from a labeled training dataset — essentially, one that already has the right answers — to inform a predictive model. As DeepAI explains it, “supervised learning is the process of teaching a model by feeding it input data as well as correct output data.”
Unsupervised learning algorithms, on the other hand, have far more limited information and, consequently, cannot be trained as in a supervised learning environment. Instead, unsupervised learning deals with the underlying infrastructure of the data and is commonly used to identify patterns and anomalies. Anomaly detection is therefore considered an unsupervised learning method.
Identifying Anomalies in Billing
Anomalies generally fall into one of three categories — point, collective or contextual.
- Point anomalies are single outliers that stand out from — or are anomalous with respect to — the rest of the data.
- Collective anomalies refer to sets of data instances that are anomalous compared to the whole data set.
- Contextual anomalies are data instances that are only anomalous in a certain context.
HMS analysts and data scientists partner with healthcare organizations to understand their business models, interpret data in context and flag anomalies that could be indicative of fraud, waste or abuse. Time series analysis is one of many techniques we employ to identify changes in data trends. We then analyze these trends using anomaly detection algorithms and decision trees to understand the context in which the anomaly occurred.
Artificial intelligence can also detect anomalies based on the inherent ability of neural networks to accurately recreate input data — whether it be text, numbers or images. As we build out these capabilities at HMS, we are currently applying alternative machine learn algorithms to provide competing methods for our clients, including statistical methods such as Gaussian, Poisson and Binomial distributions. Understanding our clients’ businesses enables us to determine the best anomaly detection algorithms for a given use case.
Future-Ready Solutions Capturing Fraud, Waste & Abuse
There are multiple unsupervised machine learning techniques that can be applied to help identify anomalies in provider billing and combat the multi-billion-dollar issue of fraud, waste and abuse. With a thorough understanding of our clients’ business models, HMS analysts can precisely identify and even anticipate the factors and incidents that can create anomalies in order to capture fraud at the source.
Drawing on thousands of data points, algorithms and machine learning technologies, HMS’ Payment Integrity solution is refreshed continuously for accuracy as well as the ability to identify emerging improper payment trends. To learn more, visit hms.com.