The internet provides an outstanding platform for individuals to easily communicate, share interests, and connect on a global scale. This connectivity fosters rich interactions and the exchange of ideas across borders. However, the same factors that enable such positive engagement also facilitate the occurrence of harassment, especially given the anonymity that the internet offers. The negative impacts of online harassment can be profound, often mirroring the harmful effects seen in in-person bullying and mobbing, leading to significant declines in mental health. To combat this issue, artificial intelligence (AI) offers a promising solution due to its capacity to quickly process and analyze extensive datasets. While AI is commonly employed with numerical data, natural language processing (NLP) techniques are specifically designed to handle textual data, making them ideal for detecting harassment in online communications. This study explores the potential benefits of integrating optimization metaheuristics with the AdaBoost classifier to improve the detection of abusive comments on YouTube. To enhance the effectiveness of this approach, a modified version of a well-established metaheuristic is introduced. The model optimized with this modified approach demonstrates promising results, achieving an accuracy of 0.904762, indicating its potential for effective harassment detection.