Depression is a prevalent mental health issue globally, affecting millions of individuals. Traditional methods for depression detection often face challenges such as subjectivity and stigma. In recent years, machine learning (ML) techniques have emerged as promising tools for objective and scalable depression detection. This paper comprehensively reviews ML methods applied to depression detection using Twitter data. Various data modalities, including text, audio, visual, and physiological data, have been explored, with text data being the focus of this study. Common ML algorithms such as logistic regression, support vector machines, decision trees, naive Bayes, k-nearest neighbors, and random forests have been employed, along with deep learning techniques like convolutional and recurrent neural networks. Evaluation metrics such as accuracy, precision, recall, and F1-score have been used to assess model performance. The study investigates the impact of demographic factors such as age and family history of depression on model performance. Overall, the results suggest that certain algorithms perform better for specific demographic categories, highlighting the need for personalized approaches in depression detection. Future research directions include exploring multimodal data fusion, deep learning models, unsupervised learning techniques, and ethical considerations to improve depression detection and mental health outcomes.