Emojis enhance textual communication by conveying emotions and providing contextual richness. This study compares the performance of supervised machine learning models such as Naive Bayes, Logistic Regression, Random Forest, K-Nearest Neighbors (KNN), and Artificial Neural Networks (ANNs) for emoji sentiment classification. A major addition in this study is the enhancement of the ANN model using an informed weight initialization technique, which speeds up convergence and reduces training time while maintaining improved performance. The experimental results showed that the Enhanced ANN (EANN) model obtained 94% accuracy, a 2% improvement over the baseline ANN model, while lowering training time from 45 to 18 units (60% decrease), highlighting the importance of initialization strategies in deep learning. The initialization method helped the EANN network avoid overfitting, resulting in increased generalization and accuracy. Proper initialization balanced the gradients during backpropagation, avoiding gradient issues that limit deep networks. Also, the informed weight initialization guaranteed that the EANN began training closer to an optimal solution, lowering the possibility of becoming confined in suboptimal local minima. The findings from this study contribute to advances in sentiment analysis and text mining, particularly in terms of improving the efficiency and accuracy of deep learning approaches.