The growth of social media over the past decade has been nothing short of phenomenal. An exponential increase has been seen on platforms like Facebook, Twitter, Instagram, LinkedIn, and YouTube, in their user base, accumulating billions of active users worldwide. Technology advancements, the ubiquitous use of smartphones, and the innate
human desire for connection don’t always contribute constructively; they might additionally end up in spreading violence in the form of bullying of others which is known as cyberbullying. As a result of cyberbullying, the victims will frequently experience anxiety, depression, and other mental diseases which can even result in suicide. Therefore the extensive need of detecting and controlling cyberbullying motivated us to automate this process. In this paper, we have introduced an approach to detect cyberbullying from social media data by using deep learning models. We have used Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network in the proposed hybrid approach along with word embedding technique called fastText. We achieved an accuracy of 91.63% with the proposed model by using a publicly available dataset containing 16,073 samples which outperformed all the state of the art models.