Comparison of K-Nearest Neighbor and Support Vector Machine Methods in Sentiment Analysis of Offline Courses

Abstract

The implementation of the New Normal policy triggers various responses from the community regarding offline learning. Some students accept this system well and have a positive view towards the adaptation of new habits. To understand public opinion on this policy, sentiment analysis is used to categorize opinions into positive or negative categories. This method is very useful in extracting opinions from social media and analyzing public responses to an issue. In sentiment analysis, two methods that are often used are K-nearest neighbor (KNN) and Support Vector Machine (SVM). KNN classifies data based on the closest distance, but this method is quite susceptible to noise. In contrast, SVM works by determining the optimal hyperplane to separate data classes, making it more stable in classification. In this research, tests were conducted using a 90:10 split data scenario. The analysis shows that the accuracy of the Support Vector Machine is higher than K-Nearest Neighbors.SVM recorded an accuracy of 63.39%, while KNN only reached 38.80%. In addition, based on performance evaluation, SVM excels in Precision, Recall, and F1-Score aspects when compared to KNN. Based on these results, it can be concluded that in sentiment analysis related to offline learning after the New Normal policy, the Support Vector Machine method is more effective than the K-Nearest Neighbor, both in terms of accuracy and overall model performance.