Fraud detection in medicare claims: A multivariate outlier detection approach
- and others
We apply traditional and customized multivariate outlier detection methods to detect fraud in medicare claims. We use two sets of 11 derived features, and one set of the 22 combined features. The features are defined so that fraudulent medicare providers should tend to have larger features values than non-fraudulent providers. Therefore we have an apriori direction ({open_quotes}large values{close_quotes}) in high dimensional feature space to search for the multivariate outliers. We focus on three issues: (1) outlier masking (Example: the presence of one outlier can make it difficult to detect a second outlier), (2) the impact of having an apriori direction to search for fraud, and (3) how to compare our detection methods. Traditional methods include Mahalanobis distances, (with and without dimension reduction), k-nearest neighbor, and density estimation methods. Some methods attempt to mitigate the outlier masking problem (for example: minimum volume ellipsoid covariance estimator). Customized methods include ranking methods (such as Spearman rank ordering) that exploit the {open_quotes}large is suspicious{close_quotes} notion. No two methods agree completely which providers are most suspicious so we present ways to compare our methods. One comparison method uses a list of known-fraudulent providers. All comparison methods restrict attention to the most suspicious providers.
- Research Organization:
- Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- Department of Defense, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 503526
- Report Number(s):
- LA-UR-97-1142; CONF-970837-1; ON: DE97005170; TRN: 97:004492
- Resource Relation:
- Conference: Knowledge discovery and data mininig, Newport Beach, CA (United States), 14-17 Aug 1997; Other Information: PBD: 1997
- Country of Publication:
- United States
- Language:
- English
Similar Records
Combining data mining and machine learning for effective user profiling
Minimizing Fraud in the Carbon Offset Market Using Blockchain Technologies