Application of predictive data mining technique to predict Claim Cost of Risk Items under Motor Class of Business: The Case of Awash Insurance Company S.C.

Abstract

The purpose of this study was to identify risk items with high claim ratio in order to take an appropriate measures during underwriting process to save profit making risk items under motor class of business. Even if the motor class of business takes the big portion of premium collection in Ethiopian insurance industry, most insurance companies indicate motor class of business as loss making line of business in their annual report. The main cause for this loss contribution is there are some risk items with high claim ratio which consumes a lot of the premium from the pool. Identifying those risk items from profit making risk item, helps a lot for an insurance company to maximize its profit. To tackle the problem of high claim cost in motor class of business, predictive data mining techniques has been employed using SVM, Naïve Bayes and Logistic Regression predictive models. The dataset used for the experiment in this study was collected from Awash insurance company specifically from underwriting and claim data tables of motor class of business. After cleaning irregularities and incomplete data in the dataset, a total of 52,831 records have been used to train the models in the ratio of 80:20. Among the used predictive models Naïve Bayes model outperformed the other two scoring 97.56% of accuracy and 98.7% precision. The challenging part of this study is lack of uniformity in conducting underwriting process. The underwriter may use either configured rate premium calculation which is similar throughout the company or flat rate which is specific to a branch and customer. This creates lack of uniformity throughout the company in terms of premium calculation. On the other hand most of records under configured premium calculation rate are complete and the values of attributes selected for this study are mandatory for the underwriter to be captured during underwriting process. Since predictive data mining techniques are aimed to identify patter of records in the dataset, only those risk item which have got insurance coverage with configured premium calculation rate in Awash insurance are included under this study. The predictive modes have been checked by new risk item as a prototype which is different from testing date and the outcome confirms the models are well trained and work correctly.