Implementation of support vector machine (SVM) based on particle swarm optimization (PSO) with synthetic minority over-sampling technique (SMOTE) on tweet data
Dina Tri Utari, Yunanda Mustofa Putri · AIP conference proceedings · 2023
In the current period, almost all individuals utilize social media to discover information.One of them is Twitter.Many details can be obtained from Twitter about technological advances, such as the launch of the Livin' by Mandiri application by Bank Mandiri.The public submitted various positive, negative, and neutral comments, especially Bank Mandiri customers, regarding the application by tweeting it on Twitter.In this study, a classification analysis was carried out on these comments using two methods, namely the Support Vector Machine (SVM) and the SVM-Particle Swarm Optimization (PSO), to get the best performance and classification model.PSO-based SVM classifier is able to optimize SVM parameters to improve classification accuracy.The dataset used consists of 4,131 positive comments, 746 negative comments, and 5,746 neutral comments.Because there is an unbalanced amount of data, a data balancing process is carried out using the Synthetic Minority Oversampling Technique (SMOTE).However, the data analyzed are only positive and negative comments to see the tendency of people to have an opinion about the Livin' by Mandiri application.Compared to SVM, SVM-PSO provides the best classification performance with an accuracy of 98.77%, the sensitivity of 99.64%, specificity of 93.57%, the precision of 98.93%, F1 score of 99.28%, and AUC of 0.966.Based on the AUC value, the model is included in the excellent classification category, meaning that the model is good in classification accuracy and predicts positive and negative comments.