PENINGKATAN PERFORMA ALGORITMAK-NEAREST NEIGBORD DALAM KELASIFIKASI DATA TIDAK SEIMBANG MENGGUNAKAN METODE SPIDER-2

Abstract

Class imbalance has become an ongoing problem in the field of Machine Learning and Classification. The group of data classes that are less known as the minority group, the other data class group is called the majority group (majority). In essence real data, data that is mined directly from the database is unbalanced. This condition makes it difficult for the classification method to perform generalization functions in the machine learning process. Almost all classification algorithms such as Naive Bayes, Decision Tree, K-Nearest Neighbor and others show very poor performance when working on data with highly unbalanced classes. The classification methods mentioned above are not equipped with the ability to deal with class imbalance problems. Many data processing methods are often used in cases of data imbalance, in this case research will be carried out using the Spider2 method. In this study, the Ecoli dataset was used, while for this study, 5 (five) different Ecoli datasets were used for each dataset for the level of data imbalance. After testing datasets with different levels of Inbalancing Ratio (IR), starting from the smallest 1.86 to 15.80, the results that explain that the KNN algorithm can improve its performance even better in terms of unbalanced data classification by adding the SPIDER- method 2 as a tool in dataset processing. In the 5 trials, the performance of the KNN algorithm can increase GM by 5.81% and FM 14.47% by adding the SPIDER-2 method to KNN.