Detection of cyberbullying in social media: Safeguarding children through AI-based digital safety mechanisms

Abstract

Background: Social networks and online platforms have become part of modern communication, opening huge opportunities for people to be exposed to the growing threat of cyberbullying. According to the 2023 studies in Sri Lanka, younger children suffered from the issue of victimization by cyberbullying more than older ones did: 13% versus 9.4%, correspondingly (12-14 years versus 15-18 years). Furthermore, it was found that cyberbullying affects 4% of the country's internet-using children across four representative provinces. With such ominous trends, the detection of cyberbullying across digital platforms stands differently and presents psychological dangers to children, including depression and suicidal tendencies. Therefore, this study aims to address past researchers' gaps and limitations in detecting cyberbullying by developing a machine learning system that can accurately detect harmful online interactions. Method: The study used a dataset of 49,783 Twitter, YouTube, and Facebook postings relevant to the cyberbullying category for training and testing. Term Frequency-Inverse Document Frequency was applied as a feature extractor, along with Principal Component Analysis, to reduce the dimension of the extracted features. Three classifications, namely Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM), were used to train the models to detect cyberbullying comments. Results: According to the model's comparison results, RF achieved the highest accuracy of 94.43%, and LR and SVM obtained an accuracy of 92.62% and 90.35%, respectively. Models used only 5000 components to train the model, which was a better approach than previous research. Conclusion: These results have marked a step toward the design of more scalable systems for detecting cyberbullying in real-time. By integrating this model into social media platforms, it can identify and remove cyberbullying comments across all comment sections as soon as they are posted. This represents one of the most effective approaches against online harassment and ensuring digital safety for children.