<![CDATA[Public complaint services are essential for improving government service quality by providing a direct channel for citizens to report issues. In Karawang Regency, the Tanggap Karawang (TANGKAR) platform serves this function; however, the manual classification of complaints causes delays and potential misrouting, especially due to the highly imbalanced distribution of complaint categories. This study develops an automatic classification model for public complaints in eight categories economy, education, health, social, infrastructure, security, environment, and transportation by integrating Term Frequency–Inverse Document Frequency (TF–IDF), Multinomial Naive Bayes, and Synthetic Minority Oversampling Technique (SMOTE). This integration addresses domain-specific class imbalance challenges, combining the computational efficiency of Naive Bayes, the feature representation strength of TF–IDF, and the improved minority class recognition from SMOTE. A dataset of 800 complaint records from TANGKAR underwent preprocessing, including cleaning, case folding, normalization, tokenizing, stemming, and stopword removal. TF–IDF with unigram and bigram features was used for feature extraction, followed by classification under two scenarios: original unbalanced data and balanced data via SMOTE. Evaluation metrics included accuracy, precision, recall, F1-score, and confusion matrix. The model achieved 85.09% accuracy without SMOTE and 83.40% with SMOTE, with notable improvement in detecting minority categories after balancing. Although overall accuracy slightly decreased, SMOTE enhanced equitable prediction across all categories. This approach advances current public complaint classification methods by adapting to the linguistic diversity and uneven category distribution in actual e-government data, supporting faster and more accurate decision-making in public complaint management systems.]]>
Copyrights © 2025
