Predictive modelling in motor caravan insurance and comparison of methods applied

Insurance is a sector that affects the financial structures of developed and developing countries positively. It appears that the insurance sector in Turkey has grown in double digits in recent years and the contribution to the Gross Domestic Product has increased. It is necessary to take advantage of the abilities of technology and machines to ensure that this situation continues at better levels. Data mining, which is used to identify invisible relationships between data, is one of the areas that need to be exploited for the development of insurance. Data mining is used in actuarial calculations and in the detection of all types of insurance fraud. Many data mining methods and algorithms are available today. Which one to use depends on the nature of the data sets. In this study, a model has been developed on motocaravan insurance, which is popular in Europe and can demonstrate the development potential in our country when necessary studies are carried out. Estimates have been made on which property persons can make this insurance. In the modeling study, the J48 algorithm that uses the decision tree technique, the algorithm in which the logistic regression is used, and the MultilayerPerceptron algorithm which uses artificial neural networks are run. The MultilayerPerceptron algorithm has the best accuracy. J48 is in second place. J48 has an accuracy ratio close to the MultilayerPerceptron algorithm. No serious success has been achieved in the logistic regression. It is thought that the results obtained will facilitate the studies to reach those who have the potential to have a motocaravan insurance. Key Words: data mining, insurance sector, motocaravan insurance, decision trees, logistic regression, artificial neural networks.