The Combinations of Fuzzy Membership Functions on Discretization in the Decision Tree-ID3 to Predict Degenerative Disease Status

Degenerative diseases are one of the leading causes of chronic disability on a global scale, significantly affecting the quality of life of sufferers. These diseases also burden the health care system and individuals financially. The implementation of preventive strategies can be postponed until an accurate prediction of the disease status can be achieved. Degenerative diseases that are the leading cause of death in many countries are coronary heart disease (CHD), while diabetes mellitus disease (DMD) increases the risk of CHD. Most of the predictor variables from the dataset to predict the status of both diseases are continuous. However, not all prediction methods, including the Decision Tree Iterative Dichotomiser3 (DTID3) method, can process continuous data. This work aims to predict the status of both degenerative diseases, CHD and DM, using the DTID3 method with continuous type predictor variables transformed using discretization with the concept of set membership. Seven prediction models using the DTID3 method are proposed to predict the status of each degenerative disease. One DTID3 model uses the concept of crisp set membership, and six DTID3 models use the concept of fuzzy set membership (FDTID3). Each prediction model of FDTID3 represents one combination of fuzzy membership functions in discretizing continuous predictor variables, and one combination consists of three membership functions. The performance of the proposed FDTID3 model depends on the fuzzy membership functions used. The hypothesis that the performance of the seven proposed models differs at least in one metric and that the performance of the FDTID3 models is higher than the DTID3 model discretized using the concept of crisp sets has been proven.