Depression is considered as the primary source of suicide and other concerns such as anxiety, memory loss, mood swings, working disability, etc. A number of modalities, data sources, and learning approaches are practiced by researchers in the past years. Therefore, this survey article focuses on identifying and exploring the amount of work performed by the researchers in relation to depression modalities, datasets, and learning techniques. 43 relevant articles from reputed journals and conferences are selected to perform this survey after proper refinement based on PRISMA standards. Concerning depression modalities, a significant number of studies showed the usage of text, images, videos, and audio data. Out of all, combination of modalities is seemed to be utilized in highest number, i.e., 37.4% due to its improved accuracy. Further, a number of datasets are obtained in literature including AVEC, Weibo, Reddit, DAIC-WoZ, Facebook, combined, and others. Due to large and easy access to data, Twitter dataset found to have high usage, i.e., 37.2% in comparison to other datasets. Further, in concern to learning techniques, DL including, RNN, CNN, LSTM, etc. are seemed to be frequently preferred by past studies as they have large ability to deal with complex patterns. Thus, in nutshell, this survey concludes that in literature, combined data from Twitter is utilized the most by employing DL techniques to effectively detect depression. Lastly, the drawbacks of the existing studies are explored that need crucial focus in future to perform improved depression analysis.