Depression, a widespread mental health disorder, affects a significant portion of the global population. Timely identification and intervention play a crucial role in ensuring effective treatment and support. Therefore, this research paper proposes a fine-tuned RoBERTa-based model for identifying depression in social media posts. In addition to the proposed model, Sentence-BERT is employed to encode social media posts into vector representations. These encoded vectors are then utilized in eight different popular classical machine learning models. The proposed fine-tuned RoBERTa model achieved a best macro F1-score of 0.55 for the development dataset and a comparable score of 0.41 for the testing dataset. Additionally, combining Sentence-BERT with Naive Bayes (S-BERT + NB) outperformed the fine-tuned RoBERTa model, achieving a slightly higher macro F1-score of 0.42. This demonstrates the effectiveness of the approach in detecting depression from social media posts.