<![CDATA[A high accuracy value in the classification process can ideally be obtained if the number of classes in the dataset is balanced. In fact, the data obtained do not all have a balanced number of classes, thus reducing the performance of the classification algorithm. In addition to the problem of an unbalanced number of classes, the attributes involved in the calculation also affect the accuracy value, so it is necessary to choose which attribute is the most influential. In this study, one method of feature selection is used, namely Forward Selection. This method is used to select which features are the most influential. SMOTE, which is one of the over-sampling algorithms, makes data with fewer classes equal to those with many classes. The results show that in the car evolution dataset the use of SMOTE can increase accuracy by 6.12% and the use of SMOTE with forward selection can increase accuracy by 6.09%. In the glass identification dataset the use of SMOTE can increase accuracy by 9.65% and the use of SMOTE with forward selection can increase accuracy by 12.6%. The use of forward selection with SMOTE is more effective for datasets that have a small number of classes.Keywords: Forward Selection, K-NN, Klasifikasi, SMOTE]]>
Copyrights © 2022
