Feature granularity for cardiac datasets using Rough Set

Abstract

Rough Set is a remarkable technique that has been successfully implemented in diverse applications including medical field. Typically, Rough Set is an efficient instrument in dealing with huge dataset in concert with missing values and granularing the features. However, large numbers of generated features reducts and rules must be chosen cautiously to reduce the processing power in dealing with massive parameters for classification. Hence, the primary objective of this study is to probe the significant reducts and rules prior to classification process of cardiac datasets from National Heart Institute (NHI), Malaysia. All-embracing analyses are presented to eradicate the insignificant attributes, reduct and rules for better classification taxonomy. Reducts with core attributes and minimal cardinality are preferred to construct new decision table, and subsequently generate high classification rates. In addition, rules with highest support, fewer length and high Rule Importance Measure (RIM) are favored since they reveal high quality performance. The results are compared in terms of the classification accuracy between the original decision table and a new decision table. It demonstrates that the rules with highest support value are more significant compared to the rules with less length.