Performance Evaluation of Naïve Bayes and SVM in Sentiment Analysis of Illegal Parking Attendants

Abstract

The increase in the number of vehicles in Indonesia has led to high demand for parking spaces, which has triggered the emergence of illegal parking attendants. This phenomenon has elicited various public responses, particularly on social media platform X. This study analyzes public sentiment toward the presence of illegal parking attendants by comparing the performance of the Naïve Bayes and Support Vector Machine (SVM) algorithms. The data used consists of 1,484 Indonesian-language tweets collected via crawling techniques. The pre-processing stage included data cleaning, case folding, word normalization, tokenization, stopword removal, and stemming. The data was then labeled with positive or negative sentiment using the InSet (Indonesia Sentiment Lexicon) approach and manually validated, before being divided into training and testing datasets. Feature extraction was performed using the TF-IDF method before being applied to the classification model. The evaluation results show that the SVM algorithm with a linear kernel approach produces the highest accuracy of 82%, outperforming Naïve Bayes: Gaussian 56%, Multinomial 74%, and Bernoulli 77%. These results are expected to contribute to the formulation of more organized and transparent parking policies, as well as demonstrate the importance of sentiment analysis as a tool to support data-driven decision making.