<p>Depression has long been a significant global health issue and remains a major challenge worldwide. Timely diagnosis and appropriate treatment have proven effective in mitigating the impact of depression. There is an urgent need for effective methods and tools to monitor mental health. Automatic depression detection methods using machine learning have attracted considerable interest from researchers. Such tools allow individuals to monitor their mental health privately, which reduces the barrier posed by social stigma. The literature presents various machine learning methods for depression detection, which can be broadly classified into two categories: first, historical records of depressive symptoms, and second, non-verbal and verbal indicators that clinicians rely on for diagnosis. This review work includes both these categories of works to help design an end-to-end machine learning pipeline, which involves multimodal data collection, preprocessing data streams, feature extraction, fusion strategies, and prediction tasks to distinguish between depressed and non-depressed subjects. This research work significantly contributes to the field of multimodal depression detection, providing useful insights that can be utilised to conduct future research in this line.</p>