Depression is a psychological phenomenon that affects the metal health of people. At the maximum level of depression, one may commit suicide. Besides, it hampers the family and social activities. As a result, identifying depressed persons is critical before proceeding with physiological treatment. Social media has become the most powerful platform for expressing feelings of individuals and also became a potential source of depressive data. Some handy research works are available for English data while it is poor for Bengali data. We have created a noble dataset in Bengali and applied different machine learning and deep learning methods for performance evaluation. A combined method of CNN and GRU gives us the highest accuracy of 87.11%. Our research finds that the performance of CNN+GRU is better compared to other previous research works.