| Identifying patients' cost category enables case management and healthcare providers to manage patients' treatments and expenditures. The objectives of this research are to select a predictive set of attributes that are associated with patients' cost categories, and to test the predictive ability of the geographical location of the patient cost category, and to develop a prediction model to identify high-, medium-, and low-cost patients using two data mining techniques: Support Vector Machines (SVM) and Backpropagation Neural Networks. The data used in this research is Medical Expenditure Panel Survey (MEPS) data for the years 2010 to 2013, it consists of answers of survey questions that reflect health conditions, employments status, insurance, financial income, spending, and more patient related information. The performance of the two approaches was compared and it was determined that the accuracy of the SVM technique reached an accuracy of 78.04% and outperformed the Backpropagation Neural Network technique which has an accuracy of 63.61%. Also, it was determined that the geographical location attribute in the dataset of this case study is not predictive to patient cost category. Some interventions were proposed in the end of this research to help healthcare providers and managers employ results of the prediction model in reducing and containing patients' costs. |
