Detecting insurance fraud using supervised and unsupervised machine learning

Fraud is a significant issue for insurance companies, generating much interest in machine learning solutions. Although supervised learning for insurance fraud detection has long been a research focus, unsupervised learning has rarely been studied in this context, and there remains insufficient evidence to guide the choice between these branches of machine learning for insurance fraud detection. Accordingly, this study evaluates supervised and unsupervised learning using proprietary insurance claim data. Furthermore, we conduct a field experiment in cooperation with an insurance company to investigate the performance of each approach in terms of identifying new fraudulent claims. We derive several important findings. Unsupervised learning, especially isolation forests, can successfully detect insurance fraud. Supervised learning also performs strongly, despite few labeled fraud cases. Interestingly, unsupervised and supervised learning detect new fraudulent claims based on different input information. Therefore, for implementation, we suggest understanding supervised and unsupervised methods as complements rather than substitutes.     

The Impact of Insurance Fraud Detection Systems

This article analyzes the impact of detection systems in an insurance fraud context. In a noncommitment Costly State Verification setting insurers can only detect fraudulent claims by performing costly audits, and policyholders are overcompensated by the optimal insurance contract. We show that auditing becomes more effective and overcompensation can be reduced, when insurers are able to condition their audits on the information provided by detection systems.     

Fraud Detection Using a Multinomial Logit Model With Missing Information

Recently, Artís, Ayuso, and Guillén (2002, Journal of Risk and Insurance 69: 325–340; henceforth AAG) estimate a logit model using claims data. Some of the claims are categorized as “honest” and other claims are known to be fraudulent. Using the approach of Hausman, Abrevaya, and Scott‐Morton (1998 Journal of Econometrics 87: 239‐269), AAG estimate a modified logit model allowing for the possibility that some claims classified as honest might actually be fraudulent. Applying this model to data on Spanish automobile insurance claims, AGG find that 5 percent of the fraudulent claims go undetected. The purpose of this article is to estimate the model of AAG using a logit model with missing information. A constrained version of this model is used to reexamine the Spanish insurance claim data. The results indicate how to identify misclassified claims. We also show how misclassified claims can be identified using the AAG approach. We show that both approaches can be used to probabilistically identify misclassified claims.     

A Multi-Line Insurance Fraud Recognition System: A Government-Led Approach in Korea

This article introduces a government-led insurance fraud detection program in Korea. The Insurance Fraud Recognition System (IFRS) uses policy and claims data from multi-lines of insurance (life, automobile, and fire), employs a three-stage statistical and link analysis to identify presumably fraudulent claims by claimant or by group, and generates system reports that the government regulator draws on to make decisions. The authors evaluate the system based on the fraud statistics and IFRS results for 2004, and offer recommendations for system improvement. This article examines existing studies about fraud, industry experiments using advanced technology, and government assistance to the insurance industry's fight against fraud in selected countries. It also provides a brief overview of the Korean insurance market, especially after the recent Asian economic crisis.     

Claims Auditing in Automobile Insurance: Fraud Detection and Deterrence Objectives

Research on insurer management of opportunism in claiming has developed in two parallel literatures. One is a theoretical literature on insurance contracting that yields predictions about the nature of optimal auditing strategies for the deterrence of fraud. The other is a literature based upon statistical analysis of claims that yields empirical strategies for the detection of fraudulent claims. This article links the two literatures by providing an empirical assessment of insurers’ auditing practices in relation to theoretical predictions. The analysis makes use of a data set on the disposition of more than 1,000 randomly selected automobile personal injury protection claims settled in the state of Massachusetts. The findings of the article are consistent with the use of rational auditing strategies by insurers and with the use of audits for both deterrence and detection.     

Deterring Fraud: The Role of General Damage Awards in Automobile Insurance Settlements

Awards for pain and suffering and other noneconomic losses account for over half of all damages awarded under third‐party auto insurance bodily injury settlements. This article hypothesizes that third‐party insurers use general damage awards to reduce the incentive to submit exaggerated claims for specific damages for injuries and lost wages. Consistent with this hypothesis, the article finds evidence using data on over 17,000 closed bodily injury claims that special damage claims that exceed their expected value receive proportionally lower general damage awards than claims that do not. Among the implications of this research is the possibility that insurers will be less zealous in challenging fraudulent special damage claims under a third‐party insurance regime than they will be under a first‐party insurance regime in which access to general damages is limited.     

Moral intensity,moral awareness and ethical predispositions: The case of insurance fraud

The issue of insurance fraud by consumers continues to perplex insurance firms, costing billions of dollars per year in the United States alone. Some analysts report that 10 per cent or more of property/casualty insurance claims are fraudulent, while less than 20 per cent of fraudulent claims being detected. Consumer attitudes are becoming more tolerant of insurance fraud in recent years. Recognizing that not all insurance fraud situations are created equal, we investigate variability in perceptions of moral intensity in dissimilar insurance padding situations in a 2 (to help others versus to profit self) × 2 (a small credit union versus a large online insurer) model and compared the results between two independent samples (college students/Millennials and an older adult population). We also investigated the impact of ethical predispositions (formalism and utilitarianism) on moral awareness and moral judgment using these four scenarios. The results suggest that the Millennials may exhibit more situationalism and more lenient judgments of collaborative versus unilateral ethical violations. In particular, ‘for self’ versus ‘for others’ comparisons show striking differences between the two age groups. The results add to the growing literature in explaining intra-personal variability in moral decision making.     

A Continuous-Time Optimal Insurance Design with Costly Monitoring

We provide a theoretical and numerical framework to study optimal insurance design under asymmetric information. We consider a continuous-time model where neither the efforts nor the outcome of an insured firm are observable to an insurer. The insured may then cause two interconnected information problems: moral hazard and fraudulent claims. We show that, when costly monitoring is available, an optimal insurance contract distinguishes the one problem from the other. Furthermore, if the insured’s downward-risk aversion is weak and if the participation constraint is not too tight, then a higher level of the monitoring technology can mitigate both problems.     

Feasible fraud and auditing probabilities for insurance companies and policyholders

Insurance claims fraud is counted among the major concerns in the insurance industry, the reason being that excess payments due to fraudulent claims account for a large percentage of the total payments each year. We formulate optimization problems from the insurance company as well as the policyholder perspective based on a costly state verification approach. In this setting??while the policyholder observes his losses privately??the insurance company can decide to verify the truthfulness of incoming claims at some cost. We show simulation results illustrating the agreement range which is characterized by all valid fraud and auditing probability combinations both stakeholders are willing to accept. Furthermore, we present the impact of different valid probability combinations on the insurance company??s and the policyholder??s objective quantities and analyze the sensitivity of the agreement range with respect to a relevant input parameter. This contribution summarizes the major findings of a working paper written by Müller et?al. (Working Papers on Risk Management and Insurance (IVW-HSG), No. 92, 2011).     

Erklärungsansätze für vertragswidriges Verhalten Versicherungsnehmern aus Sicht der ökonomischen Theorie

The two main purposes of this paper are an introduction to the economic analysis of insurance fraud and furthermore a derivation of factors that determine fraudulent behavior of policyholders on insurance markets. Consequently, we analyze the strategic decision problems of insurance companies and the policyholders and identify some factors that can help to reduce fraudulent behavior. In this context we evaluate two derived starting points for the combat against insurance fraud: fraud detection systems and a consequent charge policy of detected defrauders. We illustrate that both points can help to reduce the cost of fraud. Furthermore, we enhance our earlier analysis with respect to the empirical fact that some individuals care about fairness or — in the insurance fraud context — the legitimacy of their actions. Surprisingly, in some market situations these concerns of some policyholders do not lead to a lower fraud probability. Finally, we discuss how and to what extent insurance companies can influence such ethical concerns of policyholders. On that score, we distinguish insurance specific and insurance unspecific factors and their impact on the consumers attitudes towards insurance fraud.     

Centralizing Insurance Fraud Investigation

The study of insurance fraud and its remedy is a hot topic of research, mainly because the problem of insurance fraud is so widespread. In the United States many state governments have setup agencies to combat fraud. These Insurance Fraud Bureaus (IFB) are typically established to gather information about potential fraudulent claims, and to advise prosecuting officers on the nature of each offense. This paper presents the conditions under which more fraud will be observed in an economy where an IFB conducts all audits than in an economy where each insurance company is responsible for its own investigation. Even if fraud increases, policyholders may be better off than in economy lacking an IFB. One unambiguous case where policyholders are always better is when the IFB conducts every investigation at a cost that is equal to the industry's average.     

Fraud Classification Using Principal Component Analysis of RIDITs

This article introduces to the statistical and insurance literature a mathematical technique for an a priori classification of objects when no training sample exists for which the exact correct group membership is known. The article also provides an example of the empirical application of the methodology to fraud detection for bodily injury claims in automobile insurance. With this technique, principal component analysis of RIDIT scores (PRIDIT), an insurance fraud detector can reduce uncertainty and increase the chances of targeting the appropriate claims so that an organization will be more likely to allocate investigative resources efficiently to uncover insurance fraud. In addition, other (exogenous) empirical models can be validated relative to the PRIDIT‐derived weights for optimal ranking of fraud/nonfraud claims and/or profiling. The technique at once gives measures of the individual fraud indicator variables’ worth and a measure of individual claim file suspicion level for the entire claim file that can be used to cogently direct further fraud investigation resources. Moreover, the technique does so at a lower cost than utilizing human insurance investigators, or insurance adjusters, but with similar outcomes. More generally, this technique is applicable to other commonly encountered managerial settings in which a large number of assignment decisions are made subjectively based on ‘‘clues,‘’ which may change dramatically over time. This article explores the application of these techniques to injury insurance claims for automobile bodily injury in detail.     

Employer strategies to combat health care plan fraud

Each year health care fraud drains millions of dollars from employer-sponsored health plans. Historically, employers have taken a rather tolerant view of fraud. As the pressure to manage health plan costs increases, however, many employers are beginning to see the detection and prosecution of fraud as an appropriate part of a cost management program. Fraud in medical insurance covers a wide range of activities in terms of cost and sophistication--from misrepresenting information on a claim, to billing for services never rendered, to falsifying the existence of an entire medical organization. To complicate matters, fraudulent activities can emanate from many, many sources. Perpetrators can include employees, dependents or associates of employees, providers and employees of providers--virtually anyone able to make a claim against a plan. This article addresses actions that employers can take to reduce losses from fraud. The first section suggests policy statements and administrative procedures and guidelines that can be used to discourage employee fraud. Section two addresses the most prevalent form of fraud--provider fraud. To combat provider fraud, employers should set corporate guidelines and should enlist the assistance of employees in identifying fraudulent provider activities. Section three suggests ways to improve fraud detection through the claims payment system--often the first line of defense against fraud. Finally, section four discusses the possibility of civil and criminal remedies and reviews the legal theories under which an increasing number of fraud cases have been prosecuted.     

The Cost of Pension Insurance

This article estimates the cost of the federal pension insurance program. Pension insurance claims have an important market‐risk component, which means that the cost of the exposure cannot be estimated by discounting future claims by the risk‐free rate. Moreover, owing to the complexity of the insurance contract, its price cannot be estimated with known options formulas without introducing an error of nonquantifiable magnitude. To circumvent these problems, we model the insurance program in its full complexity and use a Monte Carlo method. By hedging the exposure with a dynamic premium policy that offloads the market risk to the insureds, one can calculate the risk‐free, or actuarial, cost of that policy. One can also characterize the nature of the subsidy and its structure across insured plans. Finally, we provide an estimate of the implicit cost of the hedge function that taxpayers currently are providing for zero remuneration. The model shows that simple contingent claims models of pension insurance result in a price that is about triple the true market cost of the insurance, and that pension insurance models that ignore market risk understate the cost by half. The solution demonstrates the broad characteristics that might characterize a credible private‐sector version of pension insurance.     

基于成本—收益理论的社会医疗保险欺诈问题研究

通过引入成本—收益理论,对社会医疗保险制度中的欺诈问题进行研究,比较该制度中各参与主体的成本、收益构成,分析各影响因子对欺诈与反欺诈行为的影响,研究表明:欺诈与反欺诈行为受行为成本、行为收益大小的制约,且有其相应的约束条件。在此基础上,提出了相应的反欺诈对策。     

Outlier detection in healthcare fraud: A case study in the Medicaid dental domain

Health care insurance fraud is a pressing problem, causing substantial and increasing costs in medical insurance programs. Due to large amounts of claims submitted, estimated at 5 billion per day, review of individual claims or providers is a difficult task. This encourages the employment of automated pre-payment controls and better post-payment decision support tools to enable subject matter expert analysis. This paper presents how to apply unsupervised outlier techniques at post-payment stage to detect fraudulent patterns of received insurance claims. A special emphasis in this paper is put on the system architecture, the metrics designed for outlier detection and the flagging of suspicious providers which may support the fraud experts in evaluating providers and reveal fraud. The algorithms were tested on Medicaid data encompassing 650,000 health-care claims and 369 dentists of one state. Two health care fraud experts evaluated flagged cases and concluded that 12 of the top 17 providers (71%) submitted suspicious claim patterns and should be referred to officials for further investigation. The remaining 5 providers (29%) could be considered mis-classifications as their patterns could be explained by special characteristics of the provider. Selecting top flagged providers is demonstrated to be a valuable as an targeting method, and individual provider analysis revealed some cases of potential fraud. The study concludes that, through outlier detection, new patterns of potential fraud can be identified and possibly utilized in future automated detection mechanisms.     

Fat-Tailed Regression Modeling with Spliced Distributions

Insurance claims data usually contain a large number of zeros and exhibits fat-tail behavior. Misestimation of one end of the tail impacts the other end of the tail of the claims distribution and can affect both the adequacy of premiums and needed reserves to hold. In addition, insured policyholders in a portfolio are naturally non-homogeneous. It is an ongoing challenge for actuaries to be able to build a predictive model that will simultaneously capture these peculiar characteristics of claims data and policyholder heterogeneity. Such models can help make improved predictions and thereby ease the decision-making process. This article proposes the use of spliced regression models for fitting insurance loss data. A primary advantage of spliced distributions is their flexibility to accommodate modeling different segments of the claims distribution with different parametric models. The threshold that breaks the segments is assumed to be a parameter, and this presents an additional challenge in the estimation. Our simulation study demonstrates the effectiveness of using multistage optimization for likelihood inference and at the same time the repercussions of model misspecification. For purposes of illustration, we consider three-component spliced regression models: the first component contains zeros, the second component models the middle segment of the loss data, and the third component models the tail segment of the loss data. We calibrate these proposed models and evaluate their performance using a Singapore auto insurance claims dataset. The estimation results show that the spliced regression model performs better than the Tweedie regression model in terms of tail fitting and prediction accuracy.     

Compound Poisson Model with Covariates

Abstract

Pet insurance in North America continues to be a growing industry. Unlike in Europe, where some countries have as much as 50% of the pet population insured, very few pets in North America are insured. Pricing practices in the past have relied on market share objectives more so than on actual experience. Pricing still continues to be performed on this basis with little consideration for actuarial principles and techniques. Developments of mortality and morbidity models to be used in the pricing model and new product development are essential for pet insurance. This paper examines insurance claims as experienced in the Canadian market. The time-to-event data are investigated using the Cox’s proportional hazards model. The claim number follows a nonhomogenous Poisson process with covariates. The claim size random variable is assumed to follow a lognormal distribution. These two models work well for aggregate claims with covariates. The first three central moments of the aggregate claims for one insured animal, as well as for a block of insured animals, are derived. We illustrate the models using data collected over an eight-year period.     

Number of Accidents or Number of Claims? An Approach with Zero‐Inflated Poisson Models for Panel Data

The  hunger for bonus  is a well-known phenomenon in insurance, meaning that the insured does not report all of his accidents to save bonus on his next year's premium. In this article, we assume that the number of accidents is based on a Poisson distribution but that the number of claims is generated by censorship of this Poisson distribution. Then, we present new models for panel count data based on the zero-inflated Poisson distribution. From the claims distributions, we propose an approximation of the accident distribution, which can provide insight into the behavior of insureds. A numerical illustration based on the reported claims of a Spanish insurance company is included to support this discussion.     

Providers’ affiliation,insurance and collusion

This paper provides a theoretical analysis of the benefits for an insurance company to develop its own network of service providers when insurance fraud is characterized by collusion between policyholders and providers. In a static framework without collusion, exclusive affiliation of providers allows insurance companies to recover some market power and to lessen competition on the insurance market. This entails a decrease in the insured’s welfare. However, exclusive affiliation of providers may entail a positive effect on customers’ surplus when insurers and providers are engaged in a repeated relationship. In particular, while insurers must cooperate to retaliate against a fraudulent provider under non-exclusive affiliation, no cooperation is needed under exclusive affiliation. In that case, an insurer is indeed able to reduce the profit of a malevolent provider by moving to collusion-proof contracts when collusion is detected, and this threat may act as a deterrent for fraudulent activities. This possibility may supplement an inefficient judicial system: it is thus a second-best optimal anti-fraud policy.