Depression has emerged as a significant public health concern, with a substantial portion of the population experiencing this mental health condition without proper recognition due to factors such as work, familial, relational, and educational pressures. The imperative need for early intervention strategies has thus become evident. Given the widespread use of social media, leveraging it for this purpose is advantageous. This study analyzed depression using deep learning techniques on social media data, specifically Twitter. A new depression tweet dataset, supported by the Circumplex Model of Affect (CMA), is created to train the deep learning models for this study. The models trained on this dataset demonstrated remarkable accuracies, with BERT and RoBERTa achieving test accuracies of over 99%, followed closely by AlBERT. The trained models are further tested on a new dataset to evaluate their efficiency, and they also performed well in detecting depression on these new data. These findings underscore the potential of social media text analysis as a valuable tool for early depression detection, ultimately contributing to the advancement of mental healthcare through timely interventions.