This paper presents a comprehensive study on the efficacy of a novel hybrid LSTM-CNN model for detecting cyberbullying in online social media text. The study evaluates the performance of the proposed model against traditional machine learning classifiers including SVM, Random Forest, and Decision Trees, using metrics such as accuracy, precision, recall, F-score, and AUC-ROC. The proposed hybrid model integrates the contextual processing capabilities of Long Short-Term Memory networks with the feature extraction proficiency of Convolutional Neural Networks, aiming to capture both the sequential and spatial dimensions of textual data. Results from the experiments demonstrate that the LSTM-CNN model significantly outperforms conventional classifiers, achieving high scores across all evaluation metrics. Additionally, ROC curve analyses further affirm the model's superior sensitivity and specificity in distinguishing between cyberbullying and non-cyberbullying instances. This research highlights the potential of deep learning approaches in enhancing the detection of cyberbullying, proposing a powerful tool for social media platforms to mitigate online harassment effectively. The findings also discuss the implications of deploying such advanced detection systems, considering the ethical dimensions of surveillance and privacy. Future directions include adapting the model to handle diverse linguistic contexts and exploring the integration of user feedback to refine classification accuracy. This study sets a precedent for the development of more sophisticated, context-aware technologies in the realm of digital safety and online community management.