Sentiment Analysis Prediction For Election Campaign In India Using Ensemble Learning And Class Balancing With Smoteenn

Abstract

In the modern digital age, consumption of social media has reached a record high. The vast majority of people share their ideas and experiences with the public using social media platforms like Google, Twitter, Facebook, YouTube, and other sites. Understanding public feeling and opinion requires a great deal of both the government & businesspeople find it important. This is the rationale behind the widespread media agencies' election-related activity in conducting different sorts of opinion polls. To analyze Indians' feelings during the election campaign, to do sentiment analysis for the election campaign, multiple questions from a developed questionnaire were used to collect data using Google Forms. Among the machine learning algorithms recommended by this study are SVM, KNN, Random Forest, Naive Bayes, Logistic Regression, Decision Tree, MLP, AdaBoost, and Ensemble Learning. Based on the Performance Evaluation of Models Using Matrices Precision, Recall, & F1 Score, decision trees and ensemble learning both obtained precision, recall, or f1 scores at 100, while Naive Bayes received the lowest scores of 57, 80, and 67 respectively, and so on. Metrics accuracy, macro-average accuracy, and weighted average accuracy are also used while evaluating the models' performance. Decision trees outperform ensemble learning as well as logistic regression in terms of precision, macro-average precision, and weighted-average precision. Overall average accuracy, average accuracy when weighted, or average accuracy for ensemble learning and logistic regression are all lower. Internationally, AdaBoost has the most accurate, the least accurate, & the average.