Depression is a prevalent mental illness that results in a sense of hopelessness and disinterest in activities. This work presents two models, each having a comprehensive methodology for predicting depression by harnessing the power of deep learning techniques. An autoencoder is used for preprocessing the text data. The autoencoder is based on the semantics of text; hence, unsupervised learning is done with K-Means clustering. The next step is implementing the baseline model on the pre-processed text data. The baseline model is a very simple feed-forward neural network with just four layers. The embedding layer, the output layer, the flattening layer, and the dense layer are the four layers. For the video model, Haar Cascade, is a classifier that is pre-trained and utilized for real-time facial recognition. Also, a pre-trained CNN model is used for depression classification. The proposed methodology lies in the Probabilistic Neural Network (PNN)-based classification. By integrating these techniques, the proposed methodology showcases promising as well as effective results in accurately identifying depression which ultimately helps in improving an individual’s mental health.