Social media platforms have become increasingly susceptible to the proliferation of automated accounts, known as Social Media Bots (SMBs). They can manipulate public discourse and disseminate misinformation. This study presents a multi-class classification approach to detect and categorize SMBs using fine-tuned transformer-based models. Through a systematic evaluation of BERT, DistilBERT, RoBERTa, DeBERTa, XLNet, and ALBERT, we investigate the effectiveness of these architectures in distinguishing between various types of social bots including spam, political, Sybil, fake accounts, and genuine human accounts. Experimental results demonstrate the superiority of our proposed approach compared to traditional machine learning and standalone deep learning methods. DistilBERT achieved the best results with an accuracy of 96.83% and a precision of 96.85%. This research contributes significantly to the field of social media bot detection and offers practical implications for enhancing online platform security.