Evaluation of Machine Learning Algorithms on Twitter Users' Sentiment Classification

This study employs sentiment analysis to evaluate the performance of various machine learning algorithms on the classification of Twitter users' opinions on the proposed redesign of Nigeria's currency, the Naira, utilising three classifiers: Recurrent Neural Network (RNN), VADER, and Naive Bayes. With a labelled dataset of 25426 tweets in total containing 10347 positive tweets, 8852 negative tweets and 6227 neutral tweets, each classifier exhibited distinct strengths and weaknesses. Naive Bayes excels in positive sentiments but shows variability across classes, RNN demonstrates notable accuracy, and VADER performs satisfactorily overall but faces challenges with sentiment complexities. The alignment of user criticism towards the Naira redesign with sentiment analysis results underscores the value of this approach in capturing public sentiment on socioeconomic policies. Beyond technical aspects, the study emphasises the role of contextual factors in sentiment analysis, providing a practical context for metrics and illuminating dynamics surrounding policy decisions like the Naira redesign. The research proposes future directions, including refining sentiment analysis frameworks, exploring time series and contextual analysis, integrating advanced NLP techniques, expanding data collection, collaborating with experts, addressing ethical considerations, and conducting long-term policy impact analyses. These recommendations aim to enhance methodological robustness, insights, and the overall impact of sentiment analysis in understanding public sentiment and its implications for economic and policy decision-making in the digital age.