PREDICTING PRESIDENTIAL ELECTIONS IN INDONESIA USING MACHINE LEARNING BASED ON CANDIDATE SOCIAL MEDIA PERFORMANCE

Entering the political year of 2024, Indonesia will conduct its first large-scale political competition because the 2024 Election is the inaugural simultaneous election in the country. Hence, predicting the election is crucial for anticipating post-election security and economic conditions. Currently, polls/surveys are widely used in Indonesia for election prediction, despite their drawbacks, such as lengthy timelines, incomplete demographic coverage, and high costs. To address these limitations, the author proposes the use of machine learning to process social media data to predict the 2024 Indonesian Presidential Election. The machine learning algorithm employed is Support Vector Regression (SVR), which, in several studies, has achieved better Mean Absolute Error (MAE), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE) values than Linear Regression. Over 6,000 posts were collected directly from the official social media profiles (Facebook, Instagram, and Twitter) of the candidates and processed along with the electability poll results from 28 survey institutions as target variables. The findings indicate that high accuracy in predicting the candidates' electability can be achieved, demonstrated by competitive or better MAE, MSE, and MAPE values compared to those of the survey institutions' polls. The use of Principal Component Analysis (PCA) in the data preparation stage marks a distinction in this research, as it provides significantly better accuracy than machine learning modeling without PCA.

Keywords: election, prediction, regression, machine learning.