The digitalization of public complaint services through the LaporGub platform in Jawa Tengah has made it easier for citizens to submit aspirations and complaints to local governments. However, the complaint classification process still faces challenges because reports are written in diverse and unstructured free-text formats. This study implements the Decision Tree algorithm to automatically classify public complaint categories in Kabupaten Blora. The dataset consists of 244 complaint records processed using the Term Frequency-Inverse Document Frequency weighting method. To reduce data distribution imbalance, several minority categories were merged, resulting in eight main classes. Model evaluation was conducted using a stratified hold-out method with an 80:20 ratio and 5-fold cross-validation. The testing results achieved an accuracy of 59.18% with a weighted F1-score of 59.79%, while cross-validation produced an average accuracy of 57.35%. In addition, feature importance analysis revealed that words such as “school,” “oil,” “fertilizer,” “agriculture,” and “road” were the dominant factors influencing the classification decisions. Based on these findings, Decision Tree demonstrates good interpretability to support the initial recommendation process for complaint categorization, although its performance is still affected by limited data size and imbalanced class distribution.
Copyrights © 2026