Handling Data Imbalance Using Text Augmentation For Classifying Public Complaints

As a public service provider, the local government is obliged to provide services for every citizen, including complaint services as a public response to the services provided. Through the E-Lapor Jogja Istimewa service of Diskominfo DIY, the public can submit complaints or problems that occur online. The classification of complaints is an integral part of accelerating the government’s response to resolving problems. Automatic classification is expected to reduce errors in the selection of complaint categories. One of the problems in classification is data imbalance which causes low classification performance. Therefore, oversampling methods are used to balance the data and improve classification performance. This research uses text augmentation-based oversampling with the Synonym Replacement technique. With the SVM method, the results show that Synonym Replacement can improve classification performance with an accuracy value of 0.84 and an f1-score of 0. S2. This result is better than the SMOTE method which produces an accuracy value of 0.77 and an f1-score of 0.70.