On social media, cyberbullying (CB) is getting more and more attention. As a result of the fame and broad utilization of online entertainment by individuals, everything being equal, it is important that web-based entertainment stages be made more secure from cyberbullying. It presents a DEA-RNN cross breed profound learning model for distinguishing CB on the Twitter web-based entertainment organization. The proposed DEA-RNN model consolidates Elman type Repetitive Brain Organizations (RNN) with a streamlined Dolphin Echolocation Calculation (DEA) to calibrate the Elman RNN's boundaries while diminishing preparation time. We completely tried DEA-RNN on a dataset of 10,000 tweets and contrasted its presentation with that of state of the art calculations like Bi-directional long momentary memory (Bi-LSTM), RNN, SVM, Multinomial Gullible Bayes (MNB), and Irregular Backwoods (RF). The exploratory outcomes show thatDEA-RNN beats any remaining strategies in all situations. It outflanked the current methodologies thought about in distinguishing CB on the Twitter stage. The DEA-RNN performed better, averaging 90.45% exactness, 89.52% accuracy, 88.98% review, 89.25% F1-score, and 90.94% explicitness.