Comparison of the Accuracy Between Naive Bayes Classifier and Support Vector Machine Algorithms for Sentiment Analysis in Mobile JKN Application Reviews

Abstract

The Mobile JKN (National Health Insurance) application is a form of BPJS Health's commitment to implementing health insurance programs since 2014. The large number of reviews of the Mobile JKN application on the Google Play Store requires sentiment analysis with an algorithm that produces the best accuracy. This research compares the accuracy obtained from the Naive Bayes Classifier (NBC) and Support Vector Machine (SVM) algorithms. This algorithm is implemented directly in sentiment analysis and combined with the Synthetic Minority Over-Sampling Technique (SMOTE) technique to overcome data imbalance. The data in this research was obtained from reviews of the Mobile JKN application on the Google Play Store using the data scraping method. We use data scraping and labeling processes before performing sentiment analysis. The sentiment analysis process includes text preprocessing and processing (modeling) by dividing the data into 30%, 40%, and 50% test data, with the rest becoming training data. The results of this research showed that the algorithm with the best accuracy was the NBC algorithm using the SMOTE technique with 50% test data and the SVM algorithm without the SMOTE technique with 50% test data. Both give the same accurate results, namely 0.90 or 90%. Experiments show that the amount of test data and the application of SMOTE affect the accuracy of the two compared algorithms.