HYBRID MACHINE LEARNING FOR CYBERBULLYING DETECTION

ABSTRACT Bullying is a pervasive issue that affects individuals of all ages and can have serious consequences in society if not effectively managed. Bullying can manifest in various forms, including physical, verbal, and cyberbullying. Among these forms, cyberbullying is particularly critical and dangerous. If left unaddressed, cyberbullying poses a risk to students' safety and can even drive them to attempt suicide or cause self-harm. Cyberbullying is more devastating than mere criticism, as it often includes false information about the target, whereas criticism cyberbullying behaviors. Moreover, existing detection methods may suffer from high false positive rates or miss instances of cyberbullying classification. In this study, a cyberbullying system is developed to identify instances of cyberbullying in online communications. Firstly, a dataset of text- based cyberbullying tweets is collected for the purpose of this study. Secondly, preprocessing techniques such as text vectorization and tokenization are employed. Thirdly, sentiment-based features are extracted from each text message using the Sentiment Intensity Analyzer (SIA). The SIA calculates sentiment scores for each message based on negative, neutral, positive, and compound sentiments. Furthermore, a Random Forest Classifier is trained on the feature matrix of the data, and a user-defined function is utilized to classify cyberbullying in a given input text typically based on truth.