Dimensionality Reduction for Improving the Performance of Risk Calculation Using Machine Learning Algorithms

Author Name(s): Vaibhav A. Hiwase, 2Dr. Avinash J. Agrawal
Author Email: hiwaseva@rknec.edu

Abstract

The profit and loss of the insured person have been mainly associated with the risk of that person. The presence of the uneven distribution of this risk defines different characteristic and lifestyle of that person. Today because of lots of information gathering from technology and ease of availability of data, underwriting and policy-making in life insurance are possible by understanding risk factors from this uneven distribution. The presence of non-risk factors in this uneven distribution only contributes to adding the noise. This noise always increases execution time though not necessarily decrease the accuracy of machine-learning models. Time can be very crucial in real time system and its importance increases in higher dimensional feature space. This paper attempts to deal with such problem in life insurance industry without compromising the accuracy significantly in any machine-learning model. The method used to focus on analyzing hidden correlation among the features taken from different characteristic and behaviour patterns of the insured person and remove unimportant noisy features, which are not playing any significant role to increase and decrease the accuracy of risk calculation. Thus, it helps to decrease execution time substantially. This will help Life Insurance Company to take the fast decision in the competitive world. In addition to this, it will protect the company from financial damage by removing noisy features for data mining.

Introduction

The terms and condition of Life Insurance policy and its popularity is a result of profit or loss and analysis in the claim of policyholders from historical data. Policy mainly decides the size and structure of payment. Hence, it became a challenging task to define a perfect size and structure of payment from which Life Insurance Company can prevent failure, make some profit and attract new customers. Although it is important to decide accurate size and structure of payment, the company may fail to do so due to adverse selection. Adverse selection in life insurance industry is a situation where the purchase of insurance policy is affected by the asymmetric information. This symmetric information creates a potential threat of policyholders claiming earlier than expected period. They became a threat when they are paying low premium payment. Therefore, there has to be some kind of measure by which we can identify this potential threat by using information from historical data and charge them high premium payment. One of the measures is a risk. Correct premium payment is generally associated with risk of a person in Life Insurance Company. This risk is nothing but the life expectancy i.e. an average time a person is expected to live. This risk in life insurance can be of ordered category. If we somehow able to understand the association of this ordered risk category with characteristic information and behaviour pattern of a person, then ordered risk will become an important dependent variable to decide accurate premium payment for the insurance industry. Thus, the problem of preventing loss can narrow down to accurate classification of the policyholder in the ordered risk category.

Conclusion

Traditionally segmentation was used in life insurance with the assumption that all individuals in the segment are “alike”. The problem was small segments suffers from lack of statistical significance and poor prediction whereas large segments may not be homogenous for decision-making. Clustering had a humble origin in life insurance and its application seems to give promising results. Today, it is important to know the behaviour of clustering in higher dimensional feature space as insurance industry needs to be more accurate to perform well in the competition. In addition, the adverse selection is a critical situation caused due to asymmetric information in the database and unshared private information from insurance buyers about their risk factors. To understand this effect and behaviour of the features in higher dimensional feature space, we cannot rely only on clustering. Proposed system removes unimportant predictors from analysis and decreases execution time of machine learning models. This will enable the firm to understand the risk factors in higher dimensional feature space so that they can be useful for accurate prediction of the risk of policyholder on an individual basis in Life Insurance Company. These risk factors can then become as the counterforce against adverse selection and useful for enhancing the loyalty of policyholder and profit as well.

810 total views, no views today

Download PDF File

About the author: admin