Document Type : Review Article

Authors

• Fa Rabbani ¹
• Behrooz Masoumi ²
• MohammadReza Keyvanpour ³

¹ Qazvin Islamic Azad University

² 2. Department of Computer and Information Technology Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran

³ 3. Department of Computer Engineering, Alzahra University, Tehran, Iran.

Abstract

Depression detection has attracted increasing attention from the data mining and machine learning communities; however, the diversity of data modalities, prediction targets, and evaluation procedures makes evidence across studies difficult to interpret. This paper presents a structured comparative review framework for organizing existing work on depression detection. Based on studies published between 2014 and 2025, the framework: (1) categorizes commonly used data sources into textual, sensor and wearable, audio-visual, and multimodal data; (2) relates major algorithmic families—traditional machine learning, deep learning, and hybrid systems—to their typical application settings; and (3) summarizes reported findings to identify recurring modality–algorithm patterns and research gaps.
Because the reviewed studies differ in datasets, sample characteristics, outcome and label definitions, and experimental designs, the reported results are interpreted descriptively rather than as a direct ranking of methods. The reviewed evidence suggests that hybrid and multimodal approaches have reported favorable metric profiles in some settings; however, their observed performance depends on fusion strategy, class distribution, outcome definition, and validation protocol. Traditional machine-learning models also remain competitive in structured settings when supported by appropriate feature selection and data preparation. The review highlights concerns related to interpretability, cross-cultural generalizability, privacy-preserving implementation, and incomplete dataset reporting. Overall, the framework clarifies how data modality, algorithmic choice, and evaluation practice interact and provides a concise basis for future depression-detection research.

Keywords

• Multimodal fusion
• Hybrid predictive models
• Mental health informatics