The Effect of Parameter Tuning and Cross Validation on Indonesian Complaint Text Classification Using the Support Vector Machine Algorithm

Abstract

Text classification aims to group text data, for example, to find some information from a large social media text dataset so that it can be used by the data owner. Manual text classification is time-consuming and difficult, so some researchers try to research text classification automatically. This study attempts to classify Indonesian text datasets using the SVM algorithm. The research was conducted in two stages, namely the first experiment without cross validation parameters and parameter tuning, then the second experiment was carried out with cross validation parameters and parameter tuning. Experiments without cross validation parameters and parameter tuning for support vector machines (SVM) obtained 89.47% accuracy with precision and recall values of 0.90 and 0.89 respectively. The second experiment used cross validation with k-5 and k-10 and tuning parameters with C constant and gamma values. Cross validation results with k-10 obtained the best accuracy with a value of 96.48% with a computation time of 40.118 seconds. Next, kernel functions in tuning parameters namely sigmoid, linear and radial basis functions are analyzed and it is found that sigmoid kernel functions achieve the best accuracy and computational time.