• Tidak ada hasil yang ditemukan

View of IMPACT OF PREDICTIVE ANALYTICS AND MACHINE LEARNING IN INSURANCE INDUSTRY IN INDIA

N/A
N/A
Protected

Academic year: 2023

Membagikan "View of IMPACT OF PREDICTIVE ANALYTICS AND MACHINE LEARNING IN INSURANCE INDUSTRY IN INDIA"

Copied!
4
0
0

Teks penuh

(1)

ACCENT JOURNAL OF ECONOMICS ECOLOGY & ENGINEERING Available Online: www.ajeee.co.in/index.php/AJEEE

Vol.03, Issue 02, February 2018, ISSN -2456-1037 (INTERNATIONAL JOURNAL) UGC APPROVED NO. 48767

1

IMPACT OF PREDICTIVE ANALYTICS AND MACHINE LEARNING IN INSURANCE INDUSTRY IN INDIA

John Ranjith R.

Department Of Management Studies, Christ (Deemed To Be University), Bengaluru.

[email protected]

AbstractBuilding a profitable insurance company has always been a challenge in India due to the negative mind-set on insurance companies. Insurance industry has long been dominated by government controlled public sector companies in both life and non-life insurance segments. This paper discusses on how insurance industry in India particularly the vehicle insurance industry could benefit from emerging technologies like predictive analytics, predictive modelling and machine learning in serving the insurer and the insured.

Keywords- Insurance in India, Vehicle Insurance, Machine Learning, Predictive modelling, Predictive analytics, risk, coverage

1.

INTRODUCTION

In India, insurance industry is at a very nascent stage. Though there are more than one billion people living in the country the majority of the people do not own even a single insurance policy.

Getting people under the insurance umbrella is a major challenge faced by both public and private sector companies.

Going by the industry trends in various sectors of the economy, a sectoral growth could happen only when there is an increase in private investment. From FY 2004 to FY 2017, the number of companies have increased from 15 to 24 in the non-life insurance space. According to the IBEF report (2017), New India Assurance leads the market with 15.18%

market share. There has also been a sizeable increase in the collection of premiums by most of the companies. An insurance is a contract signed by two parties: insurer and the insured. The insurer on receiving the filled application form specifies a certain premium to be paid for the policy period by the insured. If any claims happen when the policy is in force, claim settlement happens wherein a certain amount as determined by the insurer is given to the insured. This paper discusses the inherent issues faced by the insurers in India in the next chapter.

Then we discuss how emerging technologies like predictive analytics and machine learning could change the face of the insurance industry by benefitting all the stakeholders involved. Finally, we conclude by bringing out the advantages of using technology as a leverage in bridging the insurance coverage gap in India.

2. INHERENT ISSUES FACED BY INSURANCE SECTOR IN INDIA Though there are encouraging signs like increase in the number of insurance companies and the amount of premium collected, there are two fundamental issues plaguing the insurance industry.

The first of the two is the high premium charged by the insurers. This is an important factor as people have to pay the premium regularly for long periods of time till the policy is in force. There is also another important flaw which has been built into this premium. Premiums for policies are fixed based on the average risk involved. These are calculated mainly by using the age of the person and the type of vehicle insured. The question is:

Why does a person from a sub-urban area have to pay the same premium amount as that of the person living in a city even if they are of similar age and have similar car models? The premium calculation model averages out both and gives an amount which tends to be suitable for a person living in the city than for a person living in a sub- urban area. The risk involved in driving a car in a city is more when compared to driving outside the city limits. This is one of the main reason why companies in the insurance sector are not profitable and are not able to penetrate the rural market. The second issue is the long and hard process of claim settlement. The insurance companies are seen to be going hard on customers during the claim settlement process that it does not result in a good customer experience. And since claim settlement decreases the profitability of the companies, private sector insurance companies are not favoured by the people

(2)

ACCENT JOURNAL OF ECONOMICS ECOLOGY & ENGINEERING Available Online: www.ajeee.co.in/index.php/AJEEE

Vol.03, Issue 02, February 2018, ISSN -2456-1037 (INTERNATIONAL JOURNAL) UGC APPROVED NO. 48767

2 as they believe that the main motive of the private sector insurance players are just premiums and never a better claim settlement process. It is during the claim settlement process that the trust between the client and the insurer reaches the lowest level as there are a lot of questions that the client has to answer which might not go down well with the client. Trust is an important intangible on which the whole financial industry works and when it reaches a low level it does not auger well for the industry. Apart from these two major issues, there is the drunken driving penalties, speeding violations, dangerous driving incidents which have seen a lot of court cases still not resolved for want of enough evidence. There is always a thin line between obeying the driving rules and disobeying them. All these issues can be easily rectified and clarity could be found when technology is used. Technology has the potential to bridge all these gaps and provide a correct solution to all these issues in the insurance industry.

3. MACHINE LEARNING AND PREDICTIVE ANALYTICS TECHNIQUES USED IN INSURANCE INDUSTRY

The advent of artificial neural networks in the finance industry have been increasing of late due to the technological advancements in the computer hardware and software industry. Artificial neural networks are seen to have successfully predict bankruptcy of companies (Sharda and Wilson, 1994; Tsai and Wu, 2008;

Ribeiro et al., 2011), commodity prices (Zhang and Liao, 2014) and so on. But researches have just recently started to focus their efforts on using artificial neural networks in the insurance industry seeing the benefits that it could bring to the involved parties. All researches that have been done comparing artificial neural networks and statistical models (like Binary Logic Regression BLR and Discriminant Analysis DA) have concluded on the fact that ANN models are better equipped to handle nonlinear data. Speed, simplicity and capacity are some of the trademarks of an artificial neural network. Performance of an artificial neural network can also be improved by tuning the parameters (trial and error) unlike statistical models. Many researchers have tried to predict the

insolvency of insurance companies by using ANN models. Although these researches are similar to bankruptcy prediction it is very important to study them because the main reason for an insurance company to become insolvent is wrong prediction of claim frequency and claim severity. In 2008, Huang et al.

used feed forward neural network with back propagation algorithm on five different types of insurance policies: life, annuity, health, accident and investment oriented insurance policy. They wanted to build a single artificial neural network which could be used for all five insurance policies but came to a conclusion that one neural network for each insurance should be trained instead of training one neural network for all five types of insurances. They also have assumed that only six variables are necessary for all five insurance policies prediction. In artificial neural networks, the selection of variables plays a vital role in prediction and thus they should have chosen different variables for different insurance policies.

Another research done by Goss and Vozikis (2000) brought out some of the main disadvantages in statistical models.

In traditional approach, both dependent and independent variables have to be specified in advance which is not a good process for a dynamic insurance industry. In their research on the prediction of insolvency of life insurers they have chosen 28 insurance companies which they say as a small sample size and should be increased.

This is because the size of the training set increases the ability of a neural network for accurate prediction. They had chosen only four variables stating the reason that most financial ratios correlate with each other. They also found out that artificial neural network had less prediction error than BLR (Goss, Vozikis and George, 2000). As mentioned earlier, it is important note that selection of variables in an artificial neural network plays an important role for the success of the neural network model. In 2015, Harris and Kitchens created a genetic adaptive neural network for predicting insurance claims. They completed their research on a data comprising of 1,74,000 insurance policies and finally stated that the quality of variables could be improved. This shows that selection and quality of variables play an

(3)

ACCENT JOURNAL OF ECONOMICS ECOLOGY & ENGINEERING Available Online: www.ajeee.co.in/index.php/AJEEE

Vol.03, Issue 02, February 2018, ISSN -2456-1037 (INTERNATIONAL JOURNAL) UGC APPROVED NO. 48767

3 important factor more than the quantity of data available. Another salient feature of using artificial neural networks is the ability to do risk classification. The process of identifying different premiums for same coverage based on group characteristics is called risk classification (Yunos et al., 2016). Group characteristics in ANN (Artificial Neural Network) are denoted by the variables in selection. For different insurance policies, different variables have to be chosen which when fed into the network identifies different premiums to be set for different clientele.

While the above mentioned researches are some of the major research works done in insurance industry using machine learning not many research works have been done in India due to the unavailability of quality data available to the researchers. This might be due to low penetration of insurance industry in India and the government controlled companies having better largest market share when compared to private sector insurance companies.

4. FUTURE APPLICATION AND SUGGESTIONS

Identification and the importance of availability and selection of variables are of paramount importance in an artificial neural network for predicting the insurance claims or calculating the premiums to be charged on the insured.

Technology have grown rapidly that processing power of a system should not be a limiting factor. Since no formal theory is available for determining the optimum neural network model, the number of layers, size of the hidden layer and also the learning rate have to determine by a trial and error method by using high end processing computers for speed and capacity. By using data integration, the quality of data could be enhanced. Age of driver which could identify his experience and his previous accident history could be linked with the age of the vehicle (vehicle model year) and mileage. Mileage could bring out factors such as worn tires and loose brakes.

Global positioning system (GPS) could also be included as a variable for input to the neural network which could help the actuaries in identifying the type of area (urban, suburban, rural). More advanced techniques have evolved like collecting data from motion sensors to check

whether the driver is in a stable position while driving a car which could also reflect in the premium charged to the client.

Further research could incorporate all these variables to develop a good adaptive neural network. Such a network could drastically reduce the amount of premium collected from people living in rural area and increase the claim success ratio.

5. CONCLUSION

This article has discussed several factors that are to be taken care of while designing a proper artificial neural network for risk classification in insurance industry. It has already been concluded from various researches that artificial neural network is superior to statistical models for prediction and classification. Using the same along with proper availability and selection of variables can result in a proper tool for actuaries to use in risk classification.

REFERENCES

1.

Goss, E.P., & Vozikis, G.S. (2000).

Prediction of Insolvency of Life Insurers Through Neural Networks.

ECIS.

2.

Goss, E.P., Vozikis, G.S., & George (2000). Network Application in the Insurance Industry.

3.

Harris, T., & Kitchens, F.L. (2015).

Genetic Adaptive Neural Networks for Prediction of Insurance Claims.

4.

Huang, C., Lin, Y., & Lin, C. (2008).

Determination of Insurance Policy Using Neural Networks and Simplified Models with Factor Analysis Technique.

5.

IBEF. (2017). Insurance sector in India. Retrieved December 18, 2017, from

https://www.ibef.org/industry/insura nce-sector-india.aspx

6.

Ribeiro, M. S., Pina, J. P., Soares, J.,

& Lopes, M. C. (2011). Quantitative vs. Qualitative Criteria for Credit Risk Assessment. SSRN Electronic Journal.

doi:10.2139/ssrn.2012443

7.

Sharda, R., & Wilson, R.L. (1994).

Bankruptcy prediction using neural networks. Decision Support Systems, 11, 545-557.

8.

Tsai, C., & Wu, J. (2008). Using neural network ensembles for bankruptcy prediction and credit

(4)

ACCENT JOURNAL OF ECONOMICS ECOLOGY & ENGINEERING Available Online: www.ajeee.co.in/index.php/AJEEE

Vol.03, Issue 02, February 2018, ISSN -2456-1037 (INTERNATIONAL JOURNAL) UGC APPROVED NO. 48767

4 scoring. Expert Syst. Appl., 34, 2639- 2649.

9.

Yunos, Z. M., Ali, A., Shamsyuddin, S. M., Ismail, N., & Sallehuddin, R.

S. (2016). Predictive modeling for motor insurance claims using artificial neural networks.

International Journal of Advances in Soft Computing and its Applications, 8(3), 160-172.

10.

Zhang, Fengyi & Liao, Zhigao. (2014).

Gold Price Forecasting Based on RBF Neural Network and Hybrid Fuzzy Clustering Algorithm. Proceedings of the seventh International Conference on Management Science and Engineering Management, Lecture Notes in Electrical Engineering. 241.

73- 84. 10.1007/978-3-642-40078-0- 6.

Referensi

Dokumen terkait

Many techniques have been used to classify acute sinusitis but, in this study, the machine learning methods which includes Kernel Spherical K-Means KSPKM and Support Vector Machine SVM

As suggested by the new required courses of the program and newly developed courses, the industry needs seem to gravitate toward buyer behavior, marketing analytics, and application of