Cyberbullying Detection Using QDA and LDA Algorithms

Detecting cyberbullying is critical in today's digital landscape, where platforms like Twitter, Instagram, and TikTok are integral to communication. These platforms, expose users especially young individuals to harmful interactions, leading to severe consequences such as anxiety, depression, and even self-harm. The anonymity and global reach of the internet often embolden perpetrators, making early detection essential to prevent escalation and ensure community safety. The proposed solution improves upon earlier methods by employing Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) classifiers, which are effective for smaller, imbalanced datasets. Leveraging preprocessing techniques like stemming, tokenization, and Bag of Words ensures adaptability and efficiency. Unlike computationally intensive models like CNNs, QDA's flexibility and LDA's simplicity balance precision and recall. With accuracies of 81% (QDA) and 79% (LDA), these models provide lightweight, scalable tools for combating harmful online behavior, fostering healthier digital interactions.