Enhancing Sentiment Analysis Performance Using SMOTE and Majority Voting in Machine Learning Algorithms
Fadli Suandi, Muhammad Khairul Anam, Muhammad Firdaus, Sofiansyah Fadli, Lathifah Lathifah, Eva Yumami, Alfa Saleh, Ade Zulkarnain Hasibuan · Advances in engineering research/Advances in Engineering Research · 2024
In the digital era, sentiment analysis on social media has become increasingly important in understanding public perception of various issues.However, one of the main challenges in sentiment analysis is the issue of data imbalance, where one class (such as positive sentiment) may significantly outnumber another (such as negative or neutral sentiment).This imbalance can lead to biased predictions in machine learning models, where the majority class is favored over the minority class.To address this, Synthetic Minority Oversampling Technique (SMOTE) is used to artificially balance the dataset by creating synthetic samples from the minority class.SMOTE generates new instances by interpolating between existing minority instances, improving the distribution of the data and enhancing model performance.In this research, various machine learning algorithms are utilized to perform sentiment analysis on tweets collected with the hashtag "online learning".The SMOTE oversampling technique is applied and compared with models that do not use SMOTE.This research focuses mainly on the Majority Voting algorithm, which combines predictions from multiple models to improve overall accuracy.The test results show that using SMOTE significantly improves the model's performance, especially in terms of recall and F1-Score.The Majority Voting+SMOTE algorithm achieved the highest accuracy of 97%, demonstrating the effectiveness of this approach in handling data imbalance and producing more reliable predictions.These results confirm that SMOTE effectively improves model performance under imbalanced data conditions, especially in sentiment analysis.