Author Email: firstname.lastname@example.org
The Healthcare industry has been expanding enormously. Simultaneously, fraud in the healthcare industry is becoming a critical problem. One of the serious issues is the misuse of healthcare insurance systems. The misuse and fraud in the healthcare insurance systems significantly raise the costs of healthcare services for the providers and patients. One of the methods for healthcare fraud detection, through which these costs can be decreased, is to detect unusual medical practices which could possibly signify misuse or fraud. In this research paper, we have proposed a fraud detection approach which attempts to detect which medical practitioners show aberrant behaviour in their medical insurance claims. Based on the procedures performed by the medical practitioners, this algorithm tries to determine if, and when, medical practitioners function out of the standards of their own specialty, which may signify misuse of medical insurance systems or dearth of knowledge about billing policies. This research analyses the performance of the proposed approach against the Multinomial Naïve Bayes algorithm, used in the previous research to detect anomalies in the publicly available U.S. Medicare dataset. By analysing the results obtained by both the algorithms, we were able conclude that the proposed fraud detection approach yields better results.
Data Mining; Healthcare; Fraud Detection; Multinomial Naïve Bayes Algorithm; Link Analysis Algorithm;
Healthcare is a prominent concern for all the people in the world. Hence, for almost all the countries, healthcare has become a prime expenditure. As a result, healthcare industry has become one of the largest growing industries creating a greater impact on the countries’ economies. Healthcare industry is a compound and complicated system. It involves various specialized medical practitioners qualified to diagnose several diseases and perform various types of medical treatment plans and procedures on the patients. But in every healthcare organization, there are physicians, pharmaceutical dealers and manufacturers, and medical staff that are to be paid along with expensive medical equipment used for the treatment. As a result, every medical treatment and procedure is associated with substantial cost which is often not affordable by many patients. Hence, to make the medical treatment plans and procedures affordable to more number of people, medical insurance schemes were introduced to dispense costs across the healthcare network and so that the patients and the medical equipment can be paid uniformly. But with the growing network of the medical insurance systems, there has been a rapid rise in the misuse and fraudulent activities.
The fraud in the healthcare industry has been a perpetual and critical problem for the government and taxpayers. The healthcare fraud detection has always been a matter of interest for the researchers. Traditionally, the healthcare frauds were detected manually by the auditors who used to manually assess the medical insurance data to identify unusual and suspicious claims. The traditional methods are quite expensive and time consuming. The manual detection of frauds or misuse in the healthcare insurance systems requires immense efforts and depends on the knowledge of the domain experts which could be specious. Hence, there was a need of automated systems to detect frauds in an efficient way. The modern advances of machine learning and data mining techniques paved the way for more efficient and automated detection methods. Thus, lately, in order to innovate more novel techniques, that can detect healthcare frauds automatically, mining of healthcare insurance data has gradually become a matter of interest.
In this paper, we proposed a fraud detection approach that attempts to identify aberrant medical practitioners who have performed medical procedures that are different from the other practitioners of their field possibly indicating fraud or misuse of medical insurance systems. The proposed algorithm works on the concept of Link Analysis algorithm. We tried to analyse the performance of the proposed fraud detection algorithm against the Multinomial Naïve Bayes algorithm used in the previous study. By analysing the results of all the provider types, we conclude that, in terms of performance and accuracy, our proposed fraud detection algorithm generated better results than the Multinomial Naïve Bayes algorithm. The proposed fraud detection algorithm could detect more number of outliers or aberrant medical practitioners in less time while handling more missing values in the data than the Multinomial Naïve Bayes algorithm. A possible drawback of Naive-Bayes could be that the frequency-based probability estimate will be zero if there are no occurrences of a class label and an attribute value together. Thus, the posterior probability estimate gets affected when the probabilities are multiplied gives a zero, given a conditional independence assumption. Another drawback could be that it makes a firm assumption on the form of data distribution. It assumes that any two features are independent, which is not always the case. In this research, we executed and analysed the proposed fraud detection algorithm on the CMS dataset, which contains healthcare insurance claims data of medical practitioners for the treatment they provided to their patients. The proposed fraud detection approach considered only the attributes of medical providers’ Name, National Provider Identifier, Provider Type, HCPCS code of the procedures performed and the number of services. The CMS dataset provides other attributes such as Number of Medicare Beneficiaries, the Average Medicare Allowed Amount, Average Medicare Payment Amount, etc. which can be focussed and worked upon to detect more suspicious providers based on their insurance claims’ payment and benefits.
48 total views, 1 views today