<![CDATA[Advancement of technology make buying product/service easier because the order can be made via online. Most people will consider reviews before buying product/service, so reviews greatly affect the number of product/service purchases. Therefore, fake review classification system was created using MDLText algorithm. In this research, comparing the accuracy, precision, recall and F-measure of 4 methods namely: MDLText algorithm, and MDLText algorithm with IG feature selection, MDLText algorithm with SMOTE, MDLText algorithm with SMOTE and IG feature selection. The data used are imbalanced data with a ratio of 1 fake review to 6 original reviews. Based on the test results, for the MDLText algorithm the results obtained from the change in the parameter α with a pretty good accuracy of 49,67% but with relatively low recall of 19,35%, precision 10,53%, and F-measure 13,64%. While the use of MDLText algorithm with IG feature selection is better when the threshold is 60% in terms of accuracy that is 77,48% but in terms of precision, recall and F-measure threshold 90% has a better value of 20,94%, 100% and 34,64%. In the use of MDLText algorithm with SMOTE, the most balanced composition values are at SMOTE 200% with an accuracy value of 60.93%, recall 9.68%, precision 8,82% and F-measure 9.23%. It can be concluded that the MDLText method is better with 200% SMOTE and 60% IG threshold feature selection in terms of accuracy, precision, recall and F-measure. For the k-fold validation testing the best result for accuracy, recall and F-measure at 2 fold with 81,55%, 12,5%, 7,69% and 9,52%.]]>
Copyrights © 2020
