Ranking of indicators used for the detection of motor insurance fraud
Published: 2020, Issue 1/2020
Abstract. Since the late 1980s, researchers and corporate experts have attempted to define the indicators used for the detection of motor insurance fraud; however, to this day, no unified and widely accepted list exists. Therefore, in the present study, I have compiled a ranking of the indicators used for the detection of motor insurance fraud based on the literature, according to frequency and significance. Furthermore, based on a questionnaire-based study conducted among Romanian insurers, I have compiled a list of the most frequently used and most important indicators in Romania. Finally, I have prepared an aggregate ranking of these indicators. The results show minor differences between the aggregate ranking and the rankings constructed based on the Romanian study and the literature, respectively; thus, the aggregate ranking may serve as a good starting point for future research.
Keywords: motor insurance, insurance fraud, fraud detection, fraud indicator metrics
Introduction
Motor insurance fraud has significant implications for both the insurance sector and everyday life. In addition to reducing trust in the industry and potentially destabilising the economy, insurance fraud has a substantial impact on people's cost of living. Since insurance companies determine premiums based on their claims-payment liabilities, a higher proportion of fraudulent claims—and consequently a higher total amount of paid compensation—results in higher insurance premiums being charged.
Since the late 1980s, researchers and experts have attempted to quantify the extent of motor insurance fraud in both the United States and Western Europe. Furthermore, these researchers and experts have sought to identify indicators that may serve as signals when determining the legitimacy of an insurance claim. Nevertheless, to date, no list has been produced that would be widely accepted by both researchers and corporate experts. This is due to several reasons, the most important being that it is not even entirely clear which claims fall into the category of fraudulent demands. For example, Derrig and Ostaszewski (1995) demonstrated that a group consisting of several corporate experts might judge the same claim differently. Another important reason for the lack of a unified list is that, in scientific research (and even in corporate practice), everyone works with the indicators available to them, regardless of the actual relevance of these indicators.
The fraudulent claims arising in relation to insurance claims and their characteristics are confidential data; thus, insurance companies grant researchers access to them only in exceptional circumstances. Moreover, even within a company, only the narrowest circle of experts has access to such data, as if it were to become known (even merely through rumour) that an insurance company frequently fails to pay or pays smaller amounts of compensation to the insured (even if the reduced compensation is entirely legitimate), customers might switch insurers. For this very reason, in many cases, the insurance company will pay the requested compensation even if the given claim is suspected of being fraudulent.
An examination of the Romanian and Hungarian literature also reveals that there is no unified list of indicators used to signal fraud. Furthermore, there is practically no literature concerning motor insurance fraud in either country. Generally, while research exists in the field of financial and insurance fraud (Achim–Borlea 2020; Safta et al. 2020), regarding motor insurance fraud specifically, only descriptive articles introducing the topic and various types of fraud have been published (Gavriletea–Pleșcan 2008; Lakatos 2016), and there is a complete absence of research based on statistically grounded, sophisticated econometric analyses.
The aim of the present study is to rank the indicators used to signal motor insurance fraud.
The structure of the article is as follows: first, I briefly present the literature regarding motor insurance fraud, then I describe the indicators used to signal fraudulent insurance claims, as well as their various rankings. Subsequently, taking the presented rankings into account, I construct an aggregate ranking. Finally, I summarise the most important research findings.
A brief overview of the motor insurance fraud literature
One of the first studies concerning the indicators used in the detection of motor insurance fraud was conducted by Weisberg and Derrig (1991). The authors attempted to compile a list containing the indicators taken into account when determining the legitimacy of claims. They studied approximately 600 indemnity claims involving material damage and bodily injury from 1985, 1986, and 1989. Based on the claims examined, they established a list consisting of 65 items.
Of the 65 potential indicators of fraud, it was possible to prove that 18 indicators were statistically significant in their indicative value when determining the legitimacy of claims.
In the following years, the 65 indicators proposed by Weisberg and Derrig (1991) were examined by various studies from different perspectives. Derrig and Ostaszewski (1995), for example, pointed out that the judgements of different insurance experts, or expert groups, vary considerably. While certain experts deemed some indemnity claims legitimate, others judged these same claims to be partially or entirely illegitimate. In 1998, Weisberg and Derrig published another article (Weisberg–Derrig 1998) in which they attempted to quantify the applicability of the previously identified indicators and the effectiveness of the examination techniques. Two years later, Belhadji et al. (2000) presented a method that, through the systematic application of fraud indicators, can assist insurance companies in the fight against fraud. The authors proposed a procedure capable of selecting the most important indicators of fraud.
During the subsequent period, research into fraud indicators receded slightly, and decision support systems based on data mining and artificial intelligence emerged—that is, computer-driven algorithms and rule-based systems that can, with high probability, determine the legitimacy of an indemnity claim. Naturally, the final decision is made by an expert or an expert group, but the computer can indicate which indemnity claims merit thorough investigation by the given expert. The works of Artís et al. (1999), Artís et al. (2002), Viaene et al. (2002), Pérez et al. (2005), and Bermúdez et al. (2008) can also be categorised among such research.
Following a stagnation characteristic of the early 2010s, this type of approach based on data mining and artificial intelligence has returned to the researchers' field of interest in recent years, primarily due to the proliferation of various hybrid systems (Sundarkumar et al. 2015; Sundarkumar–Ravi 2015; Hassan–Abraham 2016; Nian et al. 2016; Subudhi–Panigrahi 2017; Wang–Xu 2018; Majhi et al. 2019).
In addition to artificial intelligence and data mining, a movement emerged in the 2000s based on the costs arising from the detection of fraudulent processes. The essence of this approach is that it approaches the entire fraud detection process from a financial perspective, examining when and in what environment it is worthwhile to investigate the legitimacy of individual indemnity claims. Furthermore, in certain cases, researchers examine whether a particular decision support system can be operated economically in real-world corporate practice. Only a few studies have been produced in this area over the past decades, the most important of which are the works of Phua et al. (2004), Viaene et al. (2007), and Zelenkov (2019).
Since the present study examines indicators used to signal fraud, I do not intend to discuss in greater detail here research related to data mining and artificial intelligence-based decision support systems, nor approaches based on the costs associated with fraud detection.
Research methodology
The aim of my research is to rank the indicators used to signal the occurrence of motor insurance fraud. Therefore, as a first step, I compiled a ranking of the indicators used in the detection of motor insurance fraud based on the literature, according to frequency and significance. I examined every study published between 1991 and 2019 that was indexed by the Web of Science database and published on the subject of motor insurance. This specifically comprises articles published in 33 journals and eight conference proceedings.
Furthermore, using a questionnaire-based study, I compiled a list of the indicators deemed most important and most frequently used in Romania. I conducted the questionnaire-based research among corporate professionals in Romania, requesting them to rank the indicators identified in the literature based on their importance, and to supplement the list with indicators they use in daily practice. The questionnaire contained 22 questions and was distributed to motor insurance providers through the Uniunea Națională a Societăților de Asigurare și Reasigurare din România2. Of the 13 relevant insurers, representatives from seven, covering 70% of the Romanian market, completed the questionnaire.
Finally, I prepared an aggregate ranking of these indicators. To achieve this, two sets of questions had to be clarified in the first step. The first concerns the weight assigned to each individual indicator. The second concerns how to eliminate the distortions caused by rankings of different lengths; for instance, while the international ranking differentiates indicators between 1 and 22, and the Romanian ranking between 1 and 28, the ranking based on significance only distinguishes between 8 positions.2
Uniunea Națională a Societăților de Asigurare și Reasigurare din România (UNSAR).
Several authors have presented the problem and the related methodology in detail (Ennas et al. 2015; Bornmann et al. 2018; Lichtenthaler 2018). In the present study, based on Bornmann et al. (2018) and Zimmermann (2013), I calculated the relative distance between the classifications of the individual rankings; that is, for each indicator, I divided the achieved rank by the maximum rank, thereby obtaining the relative score of the given indicator within that specific ranking. Subsequently, I calculated the simple arithmetic mean of the relative scores achieved by each indicator across the three rankings.
In order to verify whether all three rankings statistically significantly influence the aggregate ranking, I calculated the Kendall rank correlation between the aggregate ranking and the individual rankings. Ranking of indicators used to identify fraudulent insurance claims; Indicators used to signal fraud arising in relation to financial losses. In this chapter, I review those indicators found in the literature, as well as those monitored by experts working in the anti-fraud departments of Romanian insurance companies, which are available during the early stages of the claims process and may have signaling value when examining the legitimacy of an individual insurance claim.
Over the past decades, researchers and corporate professionals have identified and examined a very wide range of these indicators (Table 1). One of the first studies to examine potential indicators and their relevance was published in 1991 (Weisberg–Derrig 1991). The authors examined the relevance of 65 possible indicators. In this study, potential indicators regarding financial losses and bodily injuries had not yet been distinguished; however, from the perspective of indicators signalling potential fraud, this forms the basis for the rest of the decade's research.
Although authors have modified/expanded the list of indicators to a lesser extent in subsequent research, the first truly major change was published in 2004 (Phua et al. 2004). In this study, the authors examined not only indicators and methods signalling fraudulent claims but also approached the entire compensation and fraud detection process from a financial perspective, examining the cost-effectiveness of various approaches. To this end, they used only those indicators related to financial compensation claims that are already available during the initial phase of the compensation process. Due to the expansion of artificial intelligence and data mining, many studies (ten out of the 41 examined studies) used the 31 potential indicators proposed by them.3 In the following years, additional indicators and fraud detection methods were also examined; thus, 77 different indicators can be identified within the 41 scientific publications reviewed herein. The ranking of these indicators by frequency of occurrence can be observed in the second column of Table 1. In cases where the frequency of occurrence for multiple indicators is identical, I have considered that the indicators achieved the same rank. Consequently, in the ranking by literature frequency, the 77 analysed indicators received a classification between 1 and 21. At the same time, the ranking of these same indicators by significance—that is, how many times they proved to be statistically significant—can be observed in the fourth column of Table 1. I would emphasise here that certain algorithms based on data mining or artificial intelligence do not measure the significance of the indicators used. For this very reason, regarding the indicator that most frequently proves to be statistically significant (the number of previous compensation claims), it can only be said that the applied methodology proved its significance on eight occasions. Naturally, as a result, the ranking in this case varies between 1 and 8.
The final 15 rows of Table 1 contain those indicators which Romanian corporate professionals identified in their responses to my questionnaire as additional indicators that may signal potential fraudulent claims, and therefore may be taken into account when determining the legitimacy of compensation claims. The third column shows the combined ranking of the 77 indicators identified based on the literature and the additional 15 indicators suggested by Romanian corporate professionals, based on the responses of the Romanian corporate professionals.
Analysing the rankings of the indicators presented in Table 1, only minor discrepancies can be observed in most cases. For example, the indicator for accidents occurring shortly after the conclusion of the insurance contract or a few days before its expiry ranks first in the Romanian ranking, whereas it holds second place in the literature ranking. Similarly, the time elapsed between the accident and the submission of the claim ranks fifth in the Romanian ranking, while it occupies third place in the literature ranking.
The ease and low cost of accessing the database used by Phua et al. (2004) may have played an important role in the popularity of these indicators.
Table or figure – see the PDF version of the issue for its exact content.
Number of previous insurance claims
The accident occurred shortly after the conclusion of the insurance contract or a few days before its expiry The time elapsed between the accident and the submission of the claim Presence of eyewitnesses Risk classification of the accident scene Number of vehicles involved in the accident Type of insurance contract (third-party liability, CASCO, etc.) Age of the driver / insured party Issues related to the address: email address, motel, false address, living with friends, etc.
Age of the vehicle
Is there a police report regarding the accident?
Number of years since obtaining the first driving licence Excess insurance policy Type of insured vehicle: sedan, sports car, SUV, light commercial vehicle, etc.
Vehicle make
Gender of the driver
Date and time of the accident
The insured admitted that the accident was caused by their own fault Value of the vehicle Number of documents related to the claim (police report, photographs, property damage claim, driving licence, etc.) Party causing the accident: the insured person or a third party Identifier of the (internal) claims handler Marital status of the driver (single, married, widowed, divorced) New and expensive vehicle model Contradictory / differing accounts of the accident The insured is unwilling to answer questions regarding the accident / difficult to contact / uncooperative The insured is a sole trader or an employee of a family member Rented vehicle
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
A minor collision resulted in extremely high repair costs. The two drivers are relatives or friends / The insured party and the other driver share the same surname. The insured is unemployed / lives in a poor rural area. The insured obtained legal counsel very quickly. The type of accident (rear-end collision, side-impact collision, collision at a controlled intersection, unusual manoeuvre, etc.). One of the drivers involved in the accident holds foreign documents. The accident occurred at night. The vehicle is part of a commercial fleet. The insured is extremely familiar with insurance and vehicle repair jargon and/or is very well-versed in the claims procedure. The insured accepts liability for the accident too readily. The insured denies the eyewitness version of the accident. The insured is unable to produce invoices related to the vehicle repairs. The accident occurred on a weekend. Vehicle purchased with cash. Total cost of the claim. All vehicles damaged in the same accident are sent to the same repair shop. Separate insurance for accessories (specialised lighting, engine tuning, etc.). The reporter of the accident and the insurance claim: the insured or a third party. The type of service centre (manufacturer service centre, authorised service centre, local garage, etc.). The insurance agent has never seen the insured's vehicle before. The vehicle was recently purchased from abroad. Presence of a factory warranty. The number of passengers in the vehicle at the time of the accident.
Table or figure – see the PDF version of the issue for its exact content.
The insured has previously been convicted of fraud or has committed offences that suggest the possibility of fraud / The insured appears in criminal records / The insured found the vehicle reported as stolen / The vehicle was repaired before the loss adjuster could inspect it / The insured insists on receiving monetary compensation instead of repairs / The insured's income is inconsistent with the value of the vehicle / The insured is highly insistent on rapid processing / The insurance premium was paid in person and in cash / The vehicle was stolen in a shopping centre / The accident was caused by an unidentified third party / Shortly before the accident, the insured checked the extent/validity of the coverage with their agent / The insured is willing to accept a relatively small settlement instead of submitting all documentation related to the accident / False or seemingly false invoices or proofs of payment / Damages unrelated to the accident / suspicious damages / The vehicle reported as stolen was found within a short period with severe damage / During the investigation, the insured appears confused and nervous / The insured is highly aggressive (threatening to call a lawyer or contact the authorities) / The insured is experiencing personal or business financial difficulties / The number of photographs taken of the accident / The accident involved severe injuries / fatalities / A written description of the accident prepared by an expert exists / The colour of the vehicle / Type of loss adjuster (internal or external) / The past owners of the vehicle cannot be determined
Table or figure – see the PDF version of the issue for its exact content.
Naming the indicators used for signalling the ranking of indicators employed in the detection of motor insurance fraud
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
However, there are also indicators that appear at opposite ends of the two rankings. These include, for example, the presence of eyewitnesses, the risk classification of the accident scene, the number of vehicles involved in the accident, the type of insurance policy (third-party liability, CASCO, etc.), the age of the driver/insured person, the number of years since obtaining the first driving licence, false or seemingly false invoices or proof of payment, cases where a reported stolen vehicle was found within a short period with serious damage, or indicators termed non-accident-related/suspicious damage. As previously mentioned, these rankings reflect the opinions of individual researchers and corporate experts, respectively. Interestingly, however, the fact remains that Romanian experts judge certain indicators to be significantly more/less important than foreign researchers do. Unfortunately, at first glance, there appears to be no commonality among these indicators. For instance, I cannot state that these indicators primarily indicate the magnitude and nature of the loss incurred, or that they relate to the insured/driver, or perhaps to the accident scene/road conditions. For this very reason, the present study does not examine what might explain these discrepancies. In my view, this issue is too complex to be discussed in a few paragraphs, and the topic warrants an independent study focusing specifically on these differences and their causes. Nevertheless, the examples above clearly illustrate that there is no consensus regarding the importance of individual indicators. Therefore, I shall hereafter present an aggregate ranking, the aim of which is to take into account all three previously described rankings (frequency in the literature, significance, and the ranking according to corporate experts).
The aggregate ranking of indicators used to signal fraudulent insurance claims. In constructing the aggregate ranking, the 15 indicators suggested by Romanian corporate experts but not present in the literature were ranked 22nd in the international ranking, and 8th in the ranking according to significance.
Based on the relative distances between the classifications of the individual rankings, I obtained the relative score of the indicators in all three rankings; the aggregate ranking was determined by averaging these (Table 2). Table 2. The aggregate ranking of indicators used to signal fraudulent insurance claims. Names of indicators used to signal fraudulent insurance claims arising in relation to financial losses.
Table or figure – see the PDF version of the issue for its exact content.
Number of previous claims
Whether the accident occurred shortly after the conclusion of the insurance contract or a few days before its expiry. Is there a police report regarding the accident?
Risk classification of the accident scene
Time elapsed between the accident and the submission of the claim
Designation of indicators used to signal the presence of eyewitnesses. Contradictory / differing accounts of the accident. The two drivers are relatives or friends. The insured and the other driver share the same surname. Type of insurance contract (third-party liability, CASCO, etc.). Number of vehicles involved in the accident. Age of the vehicle. False or seemingly false invoices or proofs of payment. A minor collision resulted in extremely high repair costs. Age of the driver / insured. Damages unrelated to the accident / suspicious damages. Number of documents related to the claim (police report, photos, property loss claim, driving licence, etc.). The insured is unwilling to answer questions regarding the accident / difficult to contact / uncooperative. The accident occurred at night. A vehicle reported stolen was found within a short period with serious damage. Value of the vehicle. The insured accepted that the accident was caused by their own fault. Type of accident (rear-end collision, side-impact collision, collision at a controlled intersection, unusual manoeuvre, etc.). Issues regarding address: email inbox, motel, false address, living with friends, etc. The insured is extremely familiar with insurance and vehicle repair jargon and/or is very familiar with the claims procedure. During the investigation, the insured appears confused and nervous. The insured accepts responsibility for the accident too easily. Date and time of the accident. Make of the vehicle. One of the drivers involved in the accident holds foreign documents. Deductible contract. Number of years elapsed since obtaining the first driving licence. Total cost of the claim. The insurance agent has never seen the insured's vehicle before. The insured denies the eyewitnesses' version of the accident. The insured is unable to produce invoices related to the vehicle repair. Party causing the accident: the insured person or a third party. Identifier of the (internal) person processing the claim. The vehicle was recently purchased from abroad. The insured is very aggressive (threatening to call a lawyer, contacting the authorities).
Nomenclature of indicators used to denote aggregate ranking The type of insured vehicle: sedan, sport, SUV, light commercial vehicle, etc. All damaged vehicles originating from the same accident are sent to the same repair shop The vehicle is part of a commercial fleet The existence of a factory warranty The type of service centre (brand service, authorised service, local service, etc.) New and expensive vehicle model The accident occurred on a weekend Rented vehicle The number of passengers in the vehicle at the time of the accident The number of photographs taken of the accident The accident was caused by an unidentified third party The insured has previously been convicted of fraud, or has committed offences that suggest the possibility of fraud / The insured appears in criminal records The insured found the vehicle that had been reported stolen The insured is experiencing personal or business financial difficulties Shortly before the accident, the insured checked the extent/validity of the coverage with their agent Right-hand drive vehicle The gender of the driver The vehicle was repaired before the loss adjuster inspected it The repair shop identifier The marital status of the driver (single, married, widowed, divorced) The insured insists on receiving monetary compensation instead of repairs Vehicle purchased with cash The location of the accident (traffic, private parking, shopping centre parking) Separate insurance for accessories (special lighting, engine tuning, etc.) The insured's income and the value of the vehicle are not in alignment The insured is very insistent on rapid processing The insured is willing to accept a relatively small compensation amount instead of submitting all documentation related to the accident The insured is unemployed / lives in a disadvantaged area The insured obtained a lawyer very quickly The insured's vehicle performed an unexpected / inexplicable manoeuvre The reporter of the accident and the insurance claim: the insured or a third party The accident involved serious injuries / fatalities The type of reported event (collision, scratch, fire, theft, natural disaster)
Table or figure – see the PDF version of the issue for its exact content.
The insured individual or an employee of a family member The number of years elapsed since the signing of the first insurance contract with the insurance company The accident occurred in Romania or abroad The insurance premium was paid in person and in cash The driver is also the owner of the insurance contract The number of contracts signed between the insured and the insurance company (in the case of multiple vehicles) Numerous receipts related to taxi or car rentals Road conditions (dry, wet, icy) Type of insured: natural person / legal person A written description of the accident prepared by an expert exists Visibility conditions (clear sky, fog, snow, rain) Occupation of the insured Occupation of the driver Type of road (asphalt, gravel road, dirt road The vehicle was stolen in a shopping centre Colour of the vehicle Type of loss adjuster (internal or external) The previous owners of the vehicle cannot be determined The insured offers to come to the insurance office to collect the compensation rather than providing a bank account number The vehicle is still listed as the property of the previous owner
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Conclusions
In my research, I examined studies published on the topics of motor insurance fraud, as well as the detection and management of motor insurance fraud, along with the indicators found within them that serve to signal potential fraudulent activity. The first conclusion of the research is that a relatively small number of studies (41, of which eight were published in conference proceedings) and even fewer researchers/research groups have dealt in-depth with the subject. This is particularly true for the Central and Eastern European region. Another important conclusion is that the majority of this relatively limited body of research examines motor insurance fraud from the perspective of artificial intelligence and data mining, rather than from economic and/or financial perspectives. Regarding the indicators used for the signalling/detection of fraudulent insurance claims, it can be stated that no unified list exists. Every study and every individual insurance company works with its own specific list. During the present research, I established an aggregated ranking that takes into account both the indicators deemed important by the literature and the recommendations of professionals from Romanian companies. Although it does not coincide entirely with the literature-based or Romanian corporate rankings, the indicators deemed most important by the aggregated ranking hold prominent positions in all three analysed rankings. That is to say, while there may be minor discrepancies by country and by insurance company, this aggregated ranking can serve as a good starting point for future research for any researcher or corporate professional interested in the topic.
Bibliography
Achim, M. V.–Borlea, S. N. 2020. Economic and Financial Crime: Corruption, Shadow Economy, and Money Laundering. Berlin: Springer Nature.
Artís, M.–Ayuso, M.–Guillen, M. 1999. Modelling different types of automobile insurance fraud behaviour in the Spanish market. Insurance: Mathematics and Economics 24(1–2), 67–81.
Artís, M.–Ayuso, M.–Guillén, M. 2002. Detection of automobile insurance fraud with discrete choice models and misclassified claims. Journal of Risk and Insurance 69(3), 325–340.
Belhadji, E. B.–Dionne, G.–Tarkhani, F. 2000. A model for the detection of insurance fraud. The Geneva Papers on Risk and Insurance-Issues and Practice 25(4), 517–538.
Bermúdez, L.–Pérez, J. M.–Ayuso, M.–Gómez, E.–Vázquez, F. J. 2008. A Bayesian dichotomous model with asymmetric link for fraud in insurance. Insurance: Mathematics and Economics 42(2), 779–786.
Bornmann, L.–Butz, A.–Wohlrabe, K. 2018. What are the top five journals in economics? A new meta-ranking. Applied Economics 50(6), 659–675.
Derrig, R. A.–Ostaszewski, K. M. 1995. Fuzzy techniques of pattern recognition in risk and claim classification. Journal of Risk and Insurance 62(3), 447–482.
Ennas, G.–Biggio, B.–Di Guardo, M. C. 2015. Data-driven journal meta-ranking in business and management. Scientometrics 105(3), 1911–1929.
Gavriletea, M.–Plescan, M. 2008. Fraud in Motors’ insurance – Practical Aspects of Romanian Insurance Market. Theoretical and Applied Economics 12(517/supplement), 50–56.
Hassan, A. K. I.–Abraham, A. 2016. Modeling insurance fraud detection using imbalanced data classification. In: Advances in nature and biologically inspired computing.
Cham: Springer, 117–127.
Lakatos, M. 2016. Casco biztosítási csalások közgazdasági modellje. Biztosítás és Kockázat 3(2), 58–79.
Lichtenthaler, U. 2018. The world’s most innovative companies: a meta-ranking.
Journal of Strategy and Management 11(4), 497–511.
Majhi, S. K.–Bhatachharya, S.–Pradhan, R.–Biswal, S. 2019. Fuzzy clustering using salp swarm algorithm for automobile insurance fraud detection. Journal of Intelligent & Fuzzy Systems 36(3), 2333–2344.
Nian, K.–Zhang, H.–Tayal, A.–Coleman, T.–Li, Y. 2016. Auto insurance fraud detection using unsupervised spectral ranking for anomaly. The Journal of Finance and Data Science 2(1), 58–75.
Pérez, J. M.–Muguerza, J.–Arbelaitz, O.–Gurrutxaga, I.–Martín, J. I. 2005.
Consolidated tree classifier learning in a car insurance fraud detection domain with class imbalance. In: International Conference on Pattern Recognition and Image Analysis. Berlin: Springer, 381–389.
Phua, C.–Alahakoon, D.–Lee, V. 2004. Minority report in fraud detection: classification of skewed data. ACM SIGKDD Explorations Newsletter 6(1), 50–59.
Safta, I. L.–Achim, M. V.–Borlea, S. N. 2020. Manipulation of Financial Statements Through the Use of Creative Accounting. Case of Romanian Companies. Studia Universitatis „Vasile Goldis” Arad –Economics Series 30(3), 90–107.
Subudhi, S.–Panigrahi, S. 2017. Use of optimized Fuzzy C-Means clustering and supervised classifiers for automobile insurance fraud detection. Journal of King Saud University – Computer and Information Sciences 32(5), 568–575.
Sundarkumar, G. G.–Ravi, V. 2015. A novel hybrid undersampling method for mining unbalanced datasets in banking and insurance. Engineering Applications of Artificial Intelligence 37, 368–377.
Sundarkumar, G. G.–Ravi, V.–Siddeshwar, V. 2015. One-class support vector machine based undersampling: Application to churn prediction and insurance fraud detection. In: 2015 IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), 1–7.
Viaene, S.–Ayuso, M.–Guillen, M.–Van Gheel, D.–Dedene, G. 2007. Strategies for detecting fraudulent claims in the automobile insurance industry. European Journal of Operational Research 176(1), 565–583.
Viaene, S.–Derrig, R. A.–Baesens, B.–Dedene, G. 2002. A comparison of state-of-the-art classification techniques for expert automobile insurance claim fraud detection. Journal of Risk and Insurance 69(3), 373–421.
Wang, Y.–Xu, W. 2018. Leveraging deep learning with LDA-based text analytics to detect automobile insurance fraud. Decision Support Systems 105, 87–95. Weisberg, H. I.–Derrig, R. A. 1991. Fraud and Automobile Insurance: A Report on Bodily Injury Liability Claims in Massachusetts. Journal of Insurance Regulation 9(4), 497–541.
Weisberg, H. I.–Derrig, R. A. 1998. Quantitative methods for detecting fraudulent automobile bodily injury claims. Risques 35(July–September), 75–99.
Zelenkov, Y. 2019. Example-dependent cost-sensitive adaptive boosting. Expert Systems with Applications 135, 71–82.
Zimmermann, C. 2013. Academic rankings with RePEc. Econometrics 1(3), 249– 280.
A szöveget automatikusan nyertük ki a PDF-ből; a táblázatok, ábrák és a pontos tördelés a lapszám PDF-változatában láthatók.