Comparing Classification Algorithms to Analyze Twitter Sentiment on Public Opinion on Fuel Oil

Twitter is currently used as one of the applications that is often used to convey opinions or unrest by the public such as the current issue that is currently rife, namely regarding the increase in fuel oil prices (BBM) in Indonesia. With so much data scattered in a tweet that only consists of a few fragments of words, there must be valuable information or meaning that Twitter users want to convey. Text mining is a very appropriate method to do the analysis because in text mining there is a method that focuses on an opinion which will be concluded into positive and negative opinions. There are several algorithms that are most often used in data mining, namely K-Nearest Neighbor, Decision Tree, and Naive Bayes. The three algorithms used in this study were also compared with the use of feature selection Particle Swarm Optimization (PSO) which is well known for improving the performance of algorithms. There are 2165 negative opinion data and 2835 positive sentiment data. The data will be tried based on distinctive sums of data namely 1000 data, 3000 data, and 5000 data. The classic K-Nearest Neighbor algorithm obtained an accuracy result of 75.60% and made it the best when tested with 1000 data. The use of Particle Swarm Optimization (PSO) is proven to improve accuracy as when tested using 3000 data with an accuracy value of 78.21% on the K-Nearest Neighbor algorithm. In the mean time, in testing 5000 Data, the classic K-Nearest Neighbor once more gotten the most elevated exactness with a esteem of 74.02%.