Comparison of K-Nearest Neighbor And Modified K-Nearest Neighbor With Feature Selection Mutual Information And Gini Index In Informatics Journal Classsification

Abstract

With the rapid development of informatics where thousands of informatics journals have been made, a new problem has occured where grouping these journals manually has become too difficult and expensive. The writer proposes using text classification for grouping these informatics journals. This research examines the combinations of two machine learning methods, K-Nearest Neighbors (KNN) and Modified K-Nearest Neighbors with two feature selection methods, Gini Index (GI) and Mutual Information (MI) to determine the model that produces the higherst evaluation score. The data are informatics journals stored in pdf files where they are given one of 3 designated labels: Information Retrieval, Database or Others. 252 data were collected from the websites, neliti.com and garuda.ristekbrin.go.id. This research examines and compares which of the two methods, KNN and MKNN at classifying informatics journal as well as determining which combination of parameters and feature selection that produces the best result. This research finds that the combination of method and feature selection that produces the best evaluation score is MKNN with GI as feature selection producing precision score, recall score and f1-score of 97.7%