The past decade has seen an exponential increase in user-generated content on social media platforms, enabling global communication and emotional expression. Unfortunately, this has also led to a rise in toxic behavior, with users often hiding behind anonymity to engage in harmful activities. While many current techniques focus on detecting toxicity through text classification, the growing use of emojis to express emotions introduces a new layer of complexity in identifying toxic content. This study explores various machine learning methods for detecting toxicity, utilizing a dataset of toxic comments provided by Google. The best-performing model is verified by human evaluators to assess its effectiveness in real-world scenarios. In addition, we analyze the role of emojis in toxic tweets, employing data mining techniques such as frequent itemset mining and the Apriori algorithm to uncover emoji patterns associated with toxic content. To further enhance detection capabilities, we developed a website that allows users to submit content for labeling, integrating emoji sets into the classification process. Our results show that incorporating emoji patterns into the models improved detection performance, offering a more comprehensive approach to identifying toxicity in online communication.