Forecasting Election Result via Artificial Intelligence Approach: NLP and Machine Learning

Abstract

As election day approaches, politics, elections, and candidates are all topics that come up regularly in conversations. Citizens anticipate their favored candidate to be selected, and they try to predict how likely it is that their preferred candidate will be elected, as well as how likely it is that the other candidates will be elected. The goal of this article is to use NLP and machine learning algorithms to forecast the election outcome from Twitter data. For preprocessing, NLP technologies were used on the dataset. For a better result from machine learning models, punctuation and stop words were removed, lower casing, tokenization, stemming, and lemmatization were utilized. Then, using machine learning techniques such as LGBMClassifier, LogisticRegression, ExtraTreeClassifier, DecisionTreeClassifier, RandomForestClassifier, GaussianNB, and KNeighborsClassifier, each result was generated based on the input feature, which was tweet and user information, respectively. When the tweet was used as a variable, the total result was about 80% of the accuracy score, while the user information variable accounted for roughly 60% of the accuracy score. As a result of this finding, it is determined that the tweet column is a far more significant component than the user information one. Pre-trained models would be used for additional study, with the goal of getting a higher accuracy score and applying this outcome to the next Korean election.