Utomo Pujianto, Muhammad Fahmi Hidayat, Harits Ar Rosyid
Reading is an important thing for one's education development. Even so, every single person has the different reading ability, this kind of differences becomes a problem when the complexity of reading material exceeds the reader's reading ability. Therefore, it is necessary to determine the difficulty level for reading material. One of the solutions that can be applied is grouping and classifying text difficulty based on the Lexile Level. Lexile Level is a value that represents a person's reading ability and the difficulty level of reading material. In this study, there were 4 experimental scenarios. First, Lexile levels are grouped into 2 clusters, 'Difficult' and 'Easy'. Second, Lexile levels are grouped into 3 clusters, 'Difficult', 'Normal', and Easy'. Furthermore, the oversampling process is carried out on each clustering result to overcome unbalanced data distribution using SMOTE (Synthetic Minority Over-Sampling Technique). Of the four processing results, text classification is done to determine the difficulty level of reading material using Multinomial Naive Bayes. The study showed that classification on 2 clusters with oversampling produces the highest average accuracy to 84%, while classification on 2 clusters without oversampling reached 71% of average accuracy. The result of classification on 3 clusters with oversampling showed an average accuracy of 78%, and the classification on 3 clusters without oversampling showed the lowest average accuracy value, which is 53%. © 2019 IEEE.
Electrical Engineering Department, Universitas Negeri Malang, Malang, Indonesia
