Application of Machine Learning for Estimating Kenyan Motor Vehicle Insurance Premium

Abstract

Motor vehicle damage insurance is the most common type of insurance in the world and one 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 crucial to identify the main causes of motor vehicle damage, its impact on revenue for insurers and factors that contribute to 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, supported by increasing computing capabilities, data availability and the trend towards automation. Statistical regression models have numerous applications in this regard. This paper 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 aim is to identify sources of risks in motor vehicles and the variables for motor vehicle premium determination. 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 will be used to advise the insurance companies on how to charge premiums dynamically.