Fraud & Waste Pattern Recognition

Healthcare Analytics • US Government Healthcare Department

Challenge

Searching for patterns and cues in a noisy dataset is like searching the needle in a haystack. The challenge of identifying fraudulent patterns without any labels of either provider or claims was a difficult task which needed to be solved in a short time frame. The innovative methodologies we implemented with highly successful outcomes.

To showcase our consulting and big data platform capability to the US government after department, we were tasked with mining provider fraud within the Workers’ Compensation claims dataset. The data from different years varied widely and was often diverse in both volume and structure. The key challenge was to derive patterns of mal-practices by providers given the noisy data with no target labels available (Evidence of fraud).

Our Approach

We experimented with two approaches: an unsupervised approach for anomaly detection and a second supervised learning approach using decision tree (LER) obtained from Office of Inspector General (OIG) to build a model of fraudulent providers. Lists of providers found temporary or permanent removal of licenses. Features were built from the data set at the patient and provider level. Unsupervised clustering was performed to reveal potential deviant behaviour observed amongst target providers in relation to the overall population.

Results

Our work highlighted certain malpractices with respect to coding such as upcoding, unbundling etc. Deviants, providers were menus such determining high billing claims against diagnosis codes with low average claims values.

Business Value

The solution helped to highlight the advanced analytics capabilities of the team and repute trust in the big data platform, leading to continued consulting engagement with the US government department thus building a long-term trust and partnership.