Page 357 - Contributed Paper Session (CPS) - Volume 6
P. 357
CPS1966 Jessa L. S. C. et al.
Identifying preferred life insurance products
using classification trees, multinomial logistic
regression, and random forest
Jessa Luzelle S. Cuaresma, Francisco N. delos Reyes
University of the Philippines, Quezon City Philippines
Abstract
Targeting is one of the strategies implemented by marketers to achieve
efficiency in the promotion of its products. In the life insurance industry,
targeting can be useful to achieve higher penetration in the market. However,
it is still a challenge to adopt this methodology in the industry given the
limitation on the knowledge on how the available data can be utilized. This
study aims to explore the application of several predictive modelling
techniques in identifying client characteristics and behaviors that determine
their preference on life insurance products. Models that are considered are
Classification Trees (CART), Multinomial Logistic Regression (MLR), and
Random Forest (RF). Results show that while the models are not capable of
predicting minority life insurance products accurately, they are able to
generate insights on the predictor relationships which can be used by
marketers in crafting strategy for distribution, promotion, and product
development. Such insights include the preference of Unit-Linked products for
protection by an immediate family, and by the insured himself if he has high
earnings.
Keywords
Life Insurance, Unit-Linked Insurance, Insurance Marketing, Predictive
Modeling
1. Introduction
While sources of data increase and become more varied over time – from
the traditional sources such as the policy management system to the more
modern ones like health devices – insurance companies are challenged to
come up with data-driven approach to carry out their initiatives for targeting.
One method which increases its popularity in the industry is predictive
modelling. This paper will be discussing about the application of several
predictive modelling techniques in identifying client characteristics and
behaviors that determine their preference on insurance products. The results
will be used by the marketers and distributors alike to identify the products
that can be best offered to clients based on their profiles which in return can
increase their propensity to buy. In the process, predictive modelling methods
that can handle multidimensional data sets with categorical output variables
346 | I S I W S C 2 0 1 9