A Tiered Approach On Dimensional Reduction Process for Prediction of Coronary Heart Disease

Abstract

The use of dimensional reduction in the diagnostic system model of coronary heart disease, many same of case do not take into account the clinical procedures commonly used by clinicians in diagnosis. This requires that the examination be done thoroughly, thus making the high cost of diagnosis. This study aims to develop a tiered approach model in reducing dimensions for predicting CHD. The method in this research is divided into several stages, namely preprocessing, building the knowledge base and system testing. Preprocessing consists of several processes, namely the removal of missing value data, grouping attributes, and dividing data for training and testing. Knowledge base modeling is divided into three levels. The first level were the risk factor attributes, the second level were the type of chest pain & ECG, and the third were scintigraphy & coronary angiography. The knowledge base was modeled based on fuzzy rules and its inferencing process using Mamdani method. The first, fuzzy rule-based was obtained by using the FRS study. The second and third stage, using the induction rule algorithm to get the rule, then converted to fuzzy rule. The tested algorithm were C4.5, CART, and FDT. The system testing was performed by the 5-folds cross-validation method, with performance parameters based on population and individual. The test resulted using the Cleveland and Hungarian datasets, the FRS+CART combination was capable of reducing the most attributes and the highest likelihood ratio performance parameter, which was 15.96. FRS+C4.5, at least the attributes were reduced, but has an AUC performance of 80.43%, while FRS+FDT, more reduced attributes than FRS+C4.5, and AUC performance parameters are better than FRS+CART. Dimensional reduction model for prediction of CHD, capable of providing better performance than not tiered.