challenges in estimating the required amount of coverage with respect to non-life insurance policies, as there are
many factors that contribute to the changing levels of risk. For instance, in the UAE, majority of insurance
companies are continually seeking to have a robust technique to not only manage risks, but also to predict it
beforehand. In this study, the problem of classifying the risk will be addressed, particularly, property line insurance.
This research will provide many insurance companies with reliable and compelling model which can be utilized in
early stages of insurance risk estimation, leading to efficient decisions that will enhance the financial performance of
the company while reducing the risks carried.
1.1 Objectives
This research aims at constructing a risk classification model for various properties through the use of Binary
Regression. Additionally, it focuses on describing the best practices employed in the UAE insurance industry, which
will help improve the customer experience and enhance the company’s profits. Finally, it aims at successfully
minimizing potential losses by utilizing effective risk prediction models.
2. Literature Review
New business opportunities are being presented with the emergence of Internet of Things (IOT), enabling the market
to better gather data which can be used to improve the process of risk prediction in insurance industry (Baecke and
Bocca 2017). Data mining along with risk assessment methods were utilized in motor insurance, resulting in
enhanced risk control and management for the company since they were able to modify the policy conditions based
on the client’s real needs, in other words, the best way to manage risk and get more efficient results is by
customizing and integrating the insurance coverage along with the client’s usage. Additionally, this implementation
improved the speed of insurance quotation being proposed for the customer, since this method works efficiently
without the use of large historical data to predict risks. Logistic regression is the simplest form of machine learning
algorithm, having a binary dependent variable (0,1) and it is famously used for the purpose of classifying or
categorizing into binary outputs (A/B, 0/1, P/F) with the condition of having at least 2 independent variables or
predictors for the model. The predictors will be checked for how they predict while supervising their effect in the
model as well.
De Menezes et al. (2017) suggested the necessity of integrating logistic regression model with boosting to enrich the
accuracy of the model, since the logistic regression doesn’t factor in any noise in data. Therefore, expert systems are
most suitable for such complex scenarios, where boosting is utilized progressively by enforcing a classification
model to the re-weighted category of the data which is specified for training purposes. The analysis was made to
differentiate the traditional approach with the maximum likelihood model, along with the logistic regression
estimated through the boosting method for the binary classification. This was implemented on the Coronary Heart
Disease (CHD) as a function of multiple biological parameters gathered from patients, with the intention of deciding
the presence or absence of CHD. In conclusion, the results concluded that the model revealed more strength than the
traditional approach. Moreover, it performed better in terms of area under the curve, responsiveness, precision and
reduced false alarm rates. Furthermore, the application of logistic linear regression extend to building prediction
models for COVID-19 patients, as it was used to determine mortality rates in China based on the age and time taken
for triage (Josephus et al., 2021). The model demonstrated around 90% accuracy in predicting the probability of
mortality in the patients, revealing that age is the highest contributing factor for patients.
In insurance industries, the company offers indemnity if the event occurs to the insured, for a given premium to be
paid. In terms of life and health insurance, the type of risk being faced is the amount of money to be paid as a result
of an unfortunate event such as injury or death. In 2018, a study by Grant et al., (2018) proposed the use of
predictive models for enhancing risk prediction in the healthcare industry. The study implemented logistic
regression model to examine the cardiothoracic surgeries carried out, demonstrating an outline of probable risk that
could materialize and allowing for better prediction prior to its occurrence. Moreover, as the healthcare system is
profoundly dependent on the amount of money estimated and paid by insurers, it is critical to present accurate
determination for such amounts as it directly impacts the healthcare system’s performance. A recent study captured
the need for a developed model to better predict risk by putting forward a model which can detect the essential
factors in determining life insurance prices, by utilizing Neuro Fuzzy Inference System (ANFIS) to capture the non-
linear relationship between the data (Mladenovic et al., 2020). Consequently, the outcomes of the study showed that
the most influencing factor in life insurance pricing was smoking. Wang et al., (2021) used a combination of
classification along with logistic regression model in order to look into the introductory risk element for e-bike
Proceedings of the International Conference on Industrial Engineering and Operations Management
Nsukka, Nigeria, 5 - 7 April, 2022
IEOM Society International