Using data science to discover insurance fraud

Introduction

This paper aims to discuss the notion of fraud within the insurance industry. Fraud has been identified as quite prevalent within the insurance industry and can be quite costly to insurance companies. Identifying fraud can be quite difficult without the use of data and proving it even more so. This paper can be divided into three main sections. It will firstly outline what is insurance fraud and its various classifications. Secondly, it will discuss the various ways in which data can be utilised in ensuring fraudulent activities can be captured both before policy inception and during the claims process. Finally, ideas have been proposed as to how in the future we can utilise artificial intelligence and blockchains to assist in identifying data in a much more robust way.

Body

Fraudulent claims are a growing concern within the insurance industry (Hipgraves, 2013). Within the United States of America, it is estimated that between 5% to 10% of claims are considered to be fraudulent (Singer, 2019) and costs Americans at least $80 billion a year (Viaene & Dedene, 2004). Further to this, Ai, Brockett, & Golden (2009) found that due to this fraudulent activity insurance premiums could increase to $300 more due to the cost of fraud. Frogley (2015) argues that whilst it is important to discover insurance fraud and punish those that are commiting these offences, insurance companies should be trying to protect those customers who have legitimate claims and policies. Insurance companies are therefore constantly looking at new ways in which they are able to identify insurance fruad.

Insurance fraud is dynamic and can be defined in various ways. Fraud has been defined by Gill, Woolley and Gill (1994) as “knowingly making a fictitious claim, inflating a claim or adding extra items to a claim, or being in any way dishonest with the intention of gaining more than legitimate entitlement”. This can be further divided into classifications of fraud; soft or opportunistic (Kenyon & Eloff, 2017), hard or planned (Riya, 2017) and application fraud (Frogley, 2015). Soft or opportunistic fraud arises when a claimant over-inflates the size of the claim. These can often be normal honest people who see the opportunity to increase their monetary gain. They may often claim for items that were never damaged as part of the event (Kenyon & Eloff, 2017). Hard or planned fraud is generally related to criminal activity. These claimants may work individually or within organized networks and are the more serious fraudsters (Morely, Ball, & Ormerod, 2006). Lastly, application fraud is when the fraud occurs before a claim has even taken place. This is often done at the policy inception and is linked to underwriting issues. It could be that the person has lied during their inception in order to get a cheaper premium or insurance that they might not otherwise have been offered (Frogley, 2015). Due to these rising concerns of fraud within the insurance industry it is vital to have ways of combating these activities.

For many insurance companies it is their front-line staff who manage their claimant’s day to day activities. As such it is often left up to them to try and discover fraud wherever possible. The study by Morely, Ball, and Ormerod (2006) found that often these people were inexpeienced and were handed a check list of fraud indicators. This methodology of identifying fraud had complications as the staff were unable to identify anamalies from new fraud variants and instead relied only on their old fraud triggers. This issue has also been raised by Perols, Bowen, Zimmermann, and Samba (2017) who found that often fraud was treated as a homogenous event which makes fraud prediction even more difficult. The use of Big Data is becoming more common in identifying fraud both prior and during the claims process.

Big Data can be subdivided into four main dimensions: volume, velocity and variety (Kenyon & Eloff, 2017). Volume, is the amount of data that is available for the company to use. Velocity, the speed of the data processing. Variety, the various data sources that have been provided (Bologa, Bologa, & Florea, 2010). Bologa, Bologa and Florea (2010) argue that through the analysis of big data the errors that may have otherwise gone undetected due to the limitations of the human brain are able to both identified and corrected. Using Big Data the company is then able to apply the machine learning phase.

Machine learning is then divided into two sub groups; the model training phase and the accuracy testing step. During the training phase a learning algorithm should be created which has the ability to detect the useful information (Riya, 2017). This algorithm is then tested for how accurate it is able to identify fraud by using known claims that have already been identified as fruadulent. The accuracy of this alogorithm is then able to be evaluated as to how well it was able to identify the claims that are already known (Kenyon & Eloff, 2017). These models are able to be created using data science tools that are readily available such as Python, R, Hadoop and SAS. Bologa, Bologa and Florea (2010) does identify two main criticisms of the data-mining tools, the amount of data is that is made publicly available as well as the ‘lack of published well-known methods and techniques’ (p. 36). Frogley (2015) cautions that whilst better data and predictive analytics is important they should not be used independently. Using the data as well as the available front-line staff will assist in ensuring they are effectively trying to incorporate anti-fraudulent activities.

One method proposed to assist in removing the human error and improving insurance is that of Blockchain. Singer (2019) argues that the use of blockchain could significally reduce fraudulent activity within the insurance industry. A blockchain can be described as record database or a public ledger of activities that have occurred. Once the information has been entered into this system it can never be erased (Crosby, Pattanayak, Verma, & Kalyanaraman, 2016). Through a decentralized validation insurance partners can work collaboratively without colluding in identifying fraud (Singer, 2019). Perosnal information and identity can encrypted to ensure that anonymity is maintained and only identified when suscpicious activity has been noticed. However, both Singer (2019) and Gatteschi, Lamberti, Demartini, Pranteda, and Santamaria (2018) caution that blockchain technology to be used for these purposes are within their infancy and are not yet ready to be used in such way.

A more practical and available way to assist in fighting fraud is the use of artificial intelligence. Many vehicles in the modern day already have GPS, parking sensors and computer vision installed within them. Using data from these sources can assist in how risky it is to insure a customer and whether they are taking out policies honestly (Kelley, Fontanetta, Heintzman, & Pereira, 2018). With the introduction of autonomous vehicles this can then completely redefine insurance (Riikkinen, Saarijarvi, Sarlin, & Lahteenmaki, 2018). Li, Shen, and Dong (2018) investigated the way in which visual evidence can be used when assessing damages to vehicles. They were able to use data sets already available through internet search engines and then use these as comparitable examples to the images of the vehicles that were being claimed against. The images would be stored within the system and can then flag if the same images have been used again for a previous claim.

Conclusion

Insurance fraud is a very costly concern within the insurance industry. Due to new technologies available it can often be difficult to identify fraud when it is occurring. Furthermore, whilst the fraud may be identified it can often be difficult to prove that it has occurred. Currently within the insurance industry claim handlers are being used that are often inexperienced and have very basic and homogenous scripts to assist them in trying to detect fraud either during conversations with the customer or by seeing images of incidents that have occurred. Using Big Data and machine learning has been identified as one way that can assist claim managers throughout the claims process. Having already identified customers that may be suspect of fraudulent behaviour or identifying unusual activities they are able to come up with more robust ways to stop the claim from progressing. Artificial intelligence is also a growing field that will not only assist with ensuring that the truthful and innocent customers are protected but also investigating more into fraudulent behaviours. The use of blockchain has also been proposed as one idea that may eventually become more useful within the future. Whilst it may currently be in its infancy there is the potential that insurance companies can work collaboratively in ensuring fraudulent customers are not offered insurance. Whilst fraud is a concern utilising data available can ensure that these behaviours are stopped sooner rather than later.

Ai, J., Brockett, P., & Golden, L. (2009). Assessing Consumer Fraud Risk in Insurance Claims: An Unsupervised Learning Technique using Discrete and Continuous Predictor Variables. North American Actuarial Journal, 13(4), 438–458.

Bologa, A., Bologa, R., & Florea, A. (2010). Big Data and Specific Analysis Methods for Insurance Fraud Detection. Database Systems Journal, 1(1), 30–39.

Crosby, M., Pattanayak, P., Verma, S., & Kalyanaraman, V. (2016). BlockChain Technology: Beyond Bitcoin. Applied Innovation Review, 6–19.

Frogley, C. (2015). Take a holistic approach:anti-fraud programs should be impremented in each part of an insurer’s business-from product development to claims management. Best’s Review, 116(4), 72–75.

Gatteschi, V., Lamberti, F., Demartini, C., Pranteda, C., & Santamaria, V. (2018). Blockchain and Smart Contracts for Insurance: Is the technology mature enough? Future Internet, 10(2), 1–16.

Gill, K., Woolley, K., & Gill, M. (1994). Insurance Fraud: The business as a victim. In M. Gill, Crime at work. Leicester: Perpetuity Press.

Hipgraves, S. (2013). Smarter fraud investigations with big data analytics. Network Security, 7–9.

Kelley, K., Fontanetta, L., Heintzman, M., & Pereira, N. (2018). Artificial Intelligence: Implications for Social Inflation and Insurance. Risk Management and Insurance Review, 21(3), 373–387.

Kenyon, D., & Eloff, J. (2017). Big data sciene for predicting insurance claims fraud. Information Security for South Africa, 40–47.

Li, P., Shen, B., & Dong, W. (2018). An anti-fraud system for car insurance claim based on visual evidence. Ithaca:Cornell Univerity Library.

Morely, N., Ball, L., & Ormerod, T. (2006). How the detection of insurance fraud succeeds and fails. Psychology, Crime & Law, 12(2), 163–180.

Perols, J., Bowen, R., Zimmermann, C., & Samba, B. (2017). Finding Needles in a Haystack: Using Data Analytics to Improve Fraud Prediction. Accounting Review, 92(2), 221–245.

Riikkinen, M., Saarijarvi, H., Sarlin, P., & Lahteenmaki, I. (2018). Using artificial intelligence to create value in insurance. International Journal of Bank Marketing, 36(6), 1145–1168.

Riya, R. (2017). Detecting insurance claim fraud using machine learning techniques. International Conference on Circuit, Power and Computing Technologies, (pp. 1–6).

Singer, A. (2019). Can blockchain improve insurance? Risk Management, 66(1), 20–25.

Viaene, S., & Dedene, G. (2004). Insurance Fraud: Issues and Challenges. The Geneva Papers on Risk and Insurance, 29(2), 313–333.

Related posts:

Who experienced control of the aux cable used to be a continuing discussion at every house get together or atlanta divorce attorneys taxi cab; but as Bluetooth speakers become a lot more popular —…

Durante unas largas vacaciones de verano, medité sobre mis rutinas diarias, el estrés y sus efectos. Si algo tenía claro es que mi vida no tendría sentido si en otoño seguía por el camino que llevaba

This article is inspired by the proof of concept application Eternitas. It demonstrates a new way of storing, encrypting and signing documents. Both this article and the application present these…