Moh. Zulfiqar Naufal Maulana, Daniel Siahaan, Ahmad Saikhu, Evi Triandini
The growth of the software industry in Indonesia necessitates a rapid and accurate requirements analysis process. However, manually classifying software requirement descriptions is often time-consuming and highly dependent on the analyst's expertise. This study proposes an automatic classification system for software descriptions in the Indonesian language using a Natural Language Processing (NLP) approach. A primary challenge in this task is the scarcity of labeled datasets within this specific domain. To address this issue, we employ a strategy that combines a Transformer model pre-trained for the Indonesian language, namely IndoBERT, with data augmentation techniques to synthetically expand the training dataset. Three augmentation methods (Dictionary-based Synonym Replacement, Contextual Word Insertion, and Random Word Swap) were utilized to enrich the limited original dataset. The model was trained to classify descriptions into seven predefined software system categories. Experimental results show that the proposed approach achieved a robust mean Accuracy of 80.79% (σ=4.81), Precision of 84.12% (σ=1.91), Recall of 82.08% (σ=5.99), and F1-Score of 80.18% (σ=3.44), indicating stable and reliable performance. These findings demonstrate that combining a language-specific BERT model with data augmentation is an effective strategy for addressing data scarcity in domain-specific text classification tasks for the Indonesian language. © 2025 IEEE.
Institut Teknologi Sepuluh Nopember, Department of Informatics, Surabaya, Indonesia; Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia; Institut Teknologi dan Bisnis STIKOM Bali, Department of Information System, Denpasar, Indonesia
