Comparative Analysis of Deep Learning Models for Classification of Indonesian Texts

Abstract

Text classification has become one of the fundamental tasks in Natural Language Processing (NLP) due to the rapid growth of unstructured textual data generated from digital platforms. Various Deep Learning models have been developed to improve classification performance; however, their effectiveness for Indonesian text classification remains a subject of investigation because of the unique linguistic characteristics of the Indonesian language. This study aims to compare the performance of four Deep Learning models, namely Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), and IndoBERT, for Indonesian text classification. The research employed an experimental quantitative approach using an Indonesian-language text dataset obtained from publicly available sources. Data preprocessing included case folding, text cleaning, tokenization, stopword removal, stemming, and text encoding before model training. The dataset was divided into 80% training data and 20% testing data. Model performance was evaluated using Accuracy, Precision, Recall, F1-Score, and Training Time. Experimental results showed that IndoBERT achieved the best performance with an accuracy of 96.48%, precision of 96.21%, recall of 96.08%, and F1-score of 96.14%, outperforming CNN, LSTM, and BiLSTM. Although IndoBERT required a longer training time, it demonstrated superior contextual understanding of Indonesian text through its Transformer-based architecture. The findings indicate that Transformer-based models provide higher classification performance than conventional Deep Learning models for Indonesian NLP tasks. This study offers valuable insights for researchers and practitioners in selecting appropriate Deep Learning models for sentiment analysis, document classification, fake news detection, and other Indonesian text processing applications.