Abstract
Cyberbullying refers to the utilization of Social Media (SM) by individuals to engage in actions, such as humiliating, embarrassing, and defaming a target, all of which occur without any face-to-face contact. Recently, cyberflashing has emerged as an important conc ern on WhatsApp. However, previous research has neglected to address the issue of cyberflashing on SM platforms. Likewise, most of the existing works didn't identify the harmfulness of cyberbullying content. Therefore, a novel PTS-GReLU-GRU-based model for classifying cyberbullying and cyberflashing on WhatsApp, with the prediction of levels of harmfulness, is proposed in this paper. Initially, cyber flashing images are taken, which are preprocessed to enhance the image quality and to remove unwanted information. Second, human presence in the image is detected using the YOLOv3 technique. The YCbCr color model analyzes the amount of skin visible in the image. Later, the image is annotated. In the meantime, cyberbullying, offensive texts, and hate speech data are preprocessed by NLP techniques. This preprocessed data is then merged using Dice's Coefficient String similarity technique. The features are then extracted from the text and images. Thereafter, by employing I-CapSA, the best features of texts and images are selected. Likewise, the preprocessed data is given as input to the CS-Cyber BERT-based word embedding process. Eventually, cyberbullying and cyberflashing are classified with the help of a novel PTS-GReLU-GRU classifier and the level of harmfulness is predicted using the LE-ANFIS techniques. The experimental outcomes prove that the proposed model attained better accuracy and precision of 98.14% and 98.85%, respectively, thus outperforming all state-of-the-art methods.