Cyberbullying has become an important issue in modern society because of the exponential growth of social platforms online. Traditional approaches such as keyword -based filtration and manual content reviews are often unable to recognize the complexity of online interactions, especially in cases associated with satire, slang or indirect aggression. The study offers a comprehensive review of the methods of Cyberbullying detection, which begins with traditional machine learning (ML) models such as Support Vector Machine (SVM) and naive bayes, developing advanced deep learning including long-term short-term memory (LSTM) and transformer-based models as long-term). We check different extraction strategies for functions (syntactic, semantic, sensible and social characteristics), much adopted dataset (Twitter, Instagram, Formspring, Wikipedia Talk Corpus) and evaluation measures such as accuracy, precision, recall and F1 score. In addition to technical ideas, we emphasize the moral challenges of distributing these systems, including issues of justice, privacy and openness. The study concludes with open research challenges, highlights multimodal detection, multilingual adaptability, negative flexibility and the importance of the development of scalable and explanatory solutions.