The utility of stickers, GIFs and primarily emojis is profound in contemporary social media conversations. Emojis and emoticons are pictorial icons used in mainstream social media chats that amplify sentiments and express the emotions of the sender to the receiver. The use of multifarious emojis is ubiquitous in tweets of Twitter users. Each emoji expresses a sentiment which is in sync with the respective tweet’s sentiment. Thus, the problem is to formulate an equation between tweets and corresponding emoji labels. This article aims to analysing the performance of indigenous machine learning classifiers such as logistic regression, multinomial Naïve Bayes, K-nearest neighbours, support vector classifier and stochastic gradient descent on emoji prediction use cases. This chapter highlights the adverse effects of imbalanced datasets on the models’ preliminary performance. Our chapter also investigates how crucial oversampling of an imbalanced dataset is by studying the contrasting performance of each classifier before and after oversampling with synthetic minority oversampling technique. The support vector machines emoji predictor has achieved the highest accuracy of 89% post-oversampling, thus outperforming other similar works.