EMCAR: Expert Multi Class Based on Association Rule

Several experimental studies revealed that expert systems have been successfully applied in real world domains such as medical diagnoses, traffic control, and many others. However, one of the major drawbacks of classic expert systems is their reliance on human domain experts which require time, care, experience and accuracy. This shortcoming also may result in building knowledge bases that may contain inconsistent rules or contradicting rules. To treat the abovementioned we intend to propose and develop automated methods based on data mining called Associative Classification (AC) that can be easily integrated into an expert system to produce the knowledge base according to hidden correlations in the input database. The methodology employed in the proposed expert system is based on learning the rules from the database rather than inputting the rules by the knowledge engineer from the domain expert and therefore, care and accuracy as well as processing time are improved. The proposed automated expert system contains a novel learning method based on AC mining that has been evaluated on Islamic textual data according to several evaluation measures including recall, precision and classification accuracy. Furthermore, five different classification approaches: Decision trees (C4.5, KNN, SVM, MCAR and NB) and the proposed automated expert system have been tested on the Islamic data set to determine the suitable method in classifying Arabic texts.