Sentence Classification Using Machine Learning and Word Embedding: An Innovation in Indonesian Language Learning

In applied linguistics, writing assessment examines language learning. There are various genres in writing, but the evaluation always includes a syntactic component or sentence structure. This research focuses on classifying sentence structure in the Indonesian language using the Random Forest Classifier algorithm on five different experiment models, which are trained using different vectorization techniques, including bag of word (BoW), hashing, Term Frequency-Inverse Document Frequency (TF-IDF), CBoW, and skipgram vectorizers. The results showed that the accuracy of the models varied significantly, with the highest accuracy of 76% achieved by the model trained using the CBOW vectorizer. The model trained using the BoW vectorizer and skipgram vectorizer had the lowest accuracies of 65%. These results suggest that different vectorization techniques significantly impact the accuracy of the model and the CBoW vectorization technique is the most effective. While the skipgram was trained using the dataset itself before being used to vectorize the dataset, but it did not show a significant improvement in accuracy. Classifying sentence structures with various models is important and may continue to support the syntactic assessment of computer-based Indonesian language writing skills.