Abstract
With the exponential rise of social media platforms, cyber bullying has become a significant issue, requiring sophisticated techniques for effective detection and prevention. Existing machine learning approaches, while foundational, often fall short in addressing the complex and nuanced linguistic patterns inherent in cyber bullying. This paper presents a novel framework that combines Recurrent Neural Networks (RNNs) for classification and Extreme Learning Machines (ELM) for feature extraction, leveraging Deep Residual ELM (DRELM) architecture to improve accuracy. This model addresses the limitations of previous methods by enhancing the ability to capture temporal linkages and subtle language variations across social media platforms. Through experimental evaluations, the proposed framework outperforms traditional machine learning approaches, delivering superior precision, memory handling, and feature extraction capabilities. These improvements make the DRELM model a robust solution for tackling the challenging problem of cyber bullying detection.
Keywords: Cyber Bullying, Deep Residual Extreme Learning Machine, Feature Extraction, Residual Recurrent Neural Network, Social Media.