Cyberbullying Detection Using Machine Learning and Deep Learning Techniques: A Review

Abstract

The explosive growth of social media has made cyberbullying a widespread and serious concern, particularly for adolescents and young adults, with documented links to anxiety, depression, and, in extreme cases, self-harm. Because manual moderation cannot scale to the volume of content generated every day, automated detection using machine learning (ML) and deep learning (DL) has become an active research area. This paper reviews the evolution of cyberbullying detection research, from early lexicon- and rule-based methods through classical supervised classifiers such as Naïve Bayes, Support Vector Machines, and Random Forest, to modern deep architectures including Convolutional Neural Networks, Recurrent Networks, and transformer-based language models such as BERT. We summarise the typical detection pipeline, compare the strengths and weaknesses of the major algorithm families, catalogue commonly used public datasets, and discuss the open challenges that remain — including context and sarcasm understanding, multilingual and code-mixed text, multimodal (text-image-video) content, class imbalance, and real-time, cross-platform deployment. The review is intended to orient new researchers and practitioners toward promising directions for building more robust, fair, and scalable cyberbullying detection systems.