Evaluation of the Performance of Naïve Bayes and Logistic Regression in Data Analysis Sentiment Users Digital Wallet

Abstract

DANA digital wallet is application service digital finance available on Google Play Store and used For make it easier users do various transaction non - cash . Application This provide feature digital payments , money transfers, purchases credit , payment bills , as well as various transaction other through device mobile . As one of the application many digital wallets used Indonesian society , DANA obtains diverse reviews and ratings from users on the Google Play Store. Analysis sentiment done For identify opinion users to in three category , namely positive , neutral , and negative . Research data collected through the web scraping process with amount initial 1,248 reviews , then processed through case folding, cleaning, normalization stages text , tokenizing, stopword removal, and stemming before used in classification . Model evaluation is carried out against 958 reviews use accuracy, precision, recall, F1-score, and confusion matrix metrics . Research results show that Naïve Bayes obtains The accuracy value is 79.65%, macro precision 84.04 %, macro recall 63.47 %, and macro F1-score 64.61 %. Meanwhile, In addition , Logistic Regression obtained an accuracy of 88.94%, macro precision of 86.35 %, macro recall of 82.75 %, and macro F1-score of 84.19 %. Logistic Regression also produced 852 predictions . Correct of 958 reviews , more tall compared to Naïve Bayes which produced 763 predictions true . Based on results said , Logistic Regression shows better performance​ better and better balanced in classify sentiment users in the analyzed dataset . Research This expected can become reference in application of text mining and machine learning for understand opinion users as well as support evaluation quality service application digital wallet .