This systematic review examines the use of machine learning (ML) for detecting mental health (MH) disorders using passive sensing approaches, focusing on multimodal data from audio and video recordings, social media, smartphones, and wearable devices. The review aims to assess feature extraction, feature fusion, and ML methodologies applied to MH disorder detection. Key findings include varying correlations of modality-specific features influenced by demographics and personalities, the growing adoption of neural network architectures for model-level fusion, and the need for more effective ML approaches to reduce underdiagnosis. The review also highlights the importance of scalable and sophisticated ML methodologies to handle heterogeneous and extensive multimodal data. The study provides a clear taxonomy of methodological approaches and guides future researchers in selecting optimal data sources based on specific MH disorders of interest. The review is structured into sections covering research methods, data sources, feature extraction, modality fusion techniques, and ML algorithms, with a focus on passive sensing approaches due to their non-intrusive nature and better capture of natural behaviors.This systematic review examines the use of machine learning (ML) for detecting mental health (MH) disorders using passive sensing approaches, focusing on multimodal data from audio and video recordings, social media, smartphones, and wearable devices. The review aims to assess feature extraction, feature fusion, and ML methodologies applied to MH disorder detection. Key findings include varying correlations of modality-specific features influenced by demographics and personalities, the growing adoption of neural network architectures for model-level fusion, and the need for more effective ML approaches to reduce underdiagnosis. The review also highlights the importance of scalable and sophisticated ML methodologies to handle heterogeneous and extensive multimodal data. The study provides a clear taxonomy of methodological approaches and guides future researchers in selecting optimal data sources based on specific MH disorders of interest. The review is structured into sections covering research methods, data sources, feature extraction, modality fusion techniques, and ML algorithms, with a focus on passive sensing approaches due to their non-intrusive nature and better capture of natural behaviors.