부산대학교 도서관

In the field of health care, the use of data for medical insurance is a current area of research. In this report, regression models created using machine learning methods and algorithms for health insurance prediction are the object of investigation. A correlation analysis was performed on the input data, and a strong dependence was found for the features BMI and smoker. A comparative analysis was made for twenty-four models constructed using Decision Trees (DT), Support Vector Machine (SVM), Boosted, and Bagged algorithms. To evaluate the model metrics were used the coefficient of determination (R-Squared), Root Mean Square Error (RMSE) and Time for training. From the obtained experimental results, it is found that the model for the BMI feature with the Bagged algorithm has an accuracy of 0.94. The mean squared error for features Smoker and Blood Pressure of models created with the Bagged algorithm is 0.06. Models built with the Support Vector Method (SVM) require more training time than the others do. Algorithms from machine learning and statistical analysis are used to create regression models that can be useful both for health care providers and to improve the services provided.