Performance Improvement of Support Vector Machine with Firefly Algorithm for Public Complaints Classification

Public complaint services require quick responses from the relevant department. Therefore, a text classification system with high accuracy is required. One of the methods used is the Support Vector Machine (SVM). SVM is a method with better performance than other machine learning methods, but it is influenced by parameter selection and the division of training and testing data. The hyperparameter C controls the optimization between margin and classification errors. One approach to optimizing parameter selection is using the Firefly Algorithm (FFA). The data samples taken from the service were 1,209 data with 4 category classes, and the dataset was split into 50% training data, 20% validating data, and 30% testing data. We obtained the SVM-FFA test results at C = 0.269, which had the highest accuracy rate of 97.52%, with an average value in 10 trials of 96.01%. Meanwhile, the SVM test results showed the highest accuracy level of 95.87%, with an average value of 95.24%. It shows that optimizing C parameters using FFA can improve the SVM's accuracy.