Since the existence of social media, cyberbullying has been an ongoing issue globally, including in Malaysia. In today’s world, some individuals have misused technology to engage in cyberbullying, taking advantage of the anonymity provided by cyberspace. This study aims to detect cyberbully comments on YouTube, as it could bring various harms to the victims both mentally and physically. By implementing the topic modelling for cyberbully comments on YouTube, early prevention can be done to prevent the cyberbully from keep happening. Authorities and parties concerned could take legal action against those bullies as a lesson to other people to stop cyberbullying. This study implements an algorithm under NLP, which is Latent Dirichlet Allocation or LDA, as this algorithm has been widely used in topic modelling context. The dataset used for this study are the comments that have cyberbullying elements on YouTube, which have been collected up to 2713 comments. This model has gone through the pre-processing phase, where this phase has focused on transforming the raw dataset to the cleaned dataset that has been used for the trained model. Hold out method has been used for the evaluation of the model, where the dataset has been split to 80% training and 20% testing. The perplexity score has been used as the indicator of the performance for the trained model, which has recorded the value of −14.158. The result shows that LDA has generated good performance as in the case of perplexity, a lower score indicates a better model.