| dc.description.abstract | Risk models need to be estimated by insurance companies so as to predict the magnitude of claim and determine the premiums charged to the insured. This is intended to prevent losses in the future. Motor vehicle damage insurance is the most common type of insurance in the world, forming the largest sector of the insurance industry. It is also the type of insurance that generates the largest amount of loss for most insurance companies. In Kenya especially, the challenge faced by insurers is to balance the growth of the motor vehicle insurance business by increasing the customer base while also maintaining the profitability of this sector. It is therefore important to identify the main causes of problems associated with motor vehicle damage insurance, its impact on the revenue of the insurers and factors that contribute to the high motor claims to enable more accurate estimates of risk versus premium paid. In recent years the interest has increased in the use of information technology (IT) and statistical machine learning methods. This is supported by increasing computing capabilities, data availability and the trend towards automation of cumbersome or repetitive tasks. Statistical regression models have numerous applications, where they have been used in many contexts. Using a linear or generalized linear regression model in predicting insurance premiums is an area in which only a few pioneer studies have been carried out with promising results. This thesis explores applicability of new machine learning techniques such as tree-boosted models to optimize the proposed premium of prospective policy holders. It proposes two machine learning models for pricing motor vehicle damage insurance (decision trees and regression). The thesis is therefore aimed at identifying sources of risks in motor vehicles, identifying variable for motor vehicle premium determinants and then establishing a framework that will be used to regulate the premiums charged by motor vehicle insurers. Data from insurance companies has been used, which is made up of the premium rates and compensations, and other variables such as age, driver's experience, etc. Results of this thesis should be seen as successful for the use of generalized linear models in the making of car damage insurance premium rates. The established model will be used to advise the insurance companies on how to charge premiums dynamically. | en_US |
