Bullying has become more widespread in the digital age and takes the form of cyberbullying, which is particularly harmful to younger people and presents several issues because of its widespread nature and far-reaching effects. Effective detection systems are needed to address this problem, and artificial intelligence (AI) shows promise in this regard. But textual data is typically difficult for AI algorithms to handle, thus natural language processing (NLP) techniques are used to translate textual data into numerical values that AI frameworks can understand. Simultaneously, the emergence of video distribution platforms has transformed the way people consume information, but it has also encouraged negative interactions, which are particularly noticeable in comment sections. The negative effects that harassment on these platforms may have on content producers, viewers, and platforms themselves emphasize how important it is to identify harassment early on and take preventative measures. This study looks into how well the AdaBoost classifier and the term frequency-inverse document frequency (TF-IDF) technique work together to handle harassment in YouTube comments. To improve model performance, metaheuristic optimization approaches are used, and a modified version of the genetic algorithm is introduced. Additionally, rigid comparative analysis with other AdaBoost metaheuristics-tuned models was performed. The obtained results provide accuracy rates above 87%, indicating the practical feasibility of the suggested technique in promoting a positive virtual community.