This research introduces a novel approach for detecting cyberbullying in social media content by utilizing a hybrid framework that combines Bidirectional Encoder Representations from Transformers (BERT) with a Support Vector Machine (SVM) classifier. To optimize the SVM performance, its parameters were systematically tuned through a grid search strategy, resulting in superior handling of complex multiclass classification tasks. The proposed BERT-SVM ensemble achieved remarkable results, reaching an accuracy of 90% on the test dataset and significantly outperforming conventional machine learning and deep learning models such as CNN, LSTM, and standalone SVM. Furthermore, interpretability techniques, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and the BERT-based interpretability (BSRT) framework, were employed to analyze and explain the hybrid model’s predictions. This not only provided deeper insights into the model’s decision-making process but also enhanced the transparency of cyberbullying detection. Overall, the findings emphasize the effectiveness of integrating BERT embeddings with SVM to deliver accurate and interpretable multiclass cyberbullying detection.