<![CDATA[This study discusses the comparison of two optimization techniques, Grid Search and Randomized Search, in improving the accuracy of the Support Vector Machine (SVM) classification method in sentiment analysis of JKN Mobile application user reviews. The problem faced is the low accuracy of SVM classification in sentiment analysis when hyper parameters are not optimized effectively. This study aims to identify a more effective optimization technique in improving the performance of the SVM model. The research methodology includes the steps of collecting user review data, preprocessing, feature extraction, data sharing, model training, hyper parameter optimization, and model evaluation. In addition, Synthetic Minority Over-Sampling Technique (SMOTE) is used to handle class imbalance. The results show that the combination of SVM with Randomized Search provides the highest accuracy of 82%, compared to Grid Search which produces an accuracy of 81.5%. The execution time for Randomized Search is also faster, which is 1 minute 21 seconds compared to Grid Search which requires 3 minutes 28 seconds. The SMOTE technique successfully improves the balance of the distribution of negative, neutral, and positive sentiment classes, although the F1-score values for the neutral and positive classes remain low. In conclusion, Randomized Search is superior in terms of time efficiency and accuracy improvement, although challenges still remain in classifying neutral and positive sentiments.]]>
