The rise of social media platforms has fundamentally transformed how individuals interact and communicate, but as it facilitated relations between various internet users, it also precipitated an alarming increase in the spreading of racist remarks and hate-fuelled incidents. Key factors contributing to the proliferation of this harmful behavior include, among many others, the anonymity of users, digital disinhibition, and the very nature of social media with its lack of laws and restrictions, through the historical patterns of discriminatory language use, linguistic cues, and contextual analysis. This research introduces a comprehensive framework that integrates artificial intelligence, advanced natural language processing tools, and machine learning classifiers to identify racist content. Leveraging these methodologies and incorporating preprocessing techniques to reduce noise and biases present in social media data leads to more accurate identification of racially objectionable content. The demonstrative outcomes show the efficacy of this approach in identifying and classifying racist remarks on Twitter, supporting automatic content moderation, and encouraging a more welcoming online community. An additional contribution of this work is the introduction of a modified metaheuristics optimization method that is applied to classifier optimization to ensure favorable outcomes. A comparative analysis is conducted on real-world data and the best-performing optimized classifiers demonstrate an accuracy of 0.923058.