A Comprehensive Survey on Human Depression Detection and Classification Using Machine learning

Abstract

Depression is a prevalent mental health disorder affecting millions globally, with serious consequences for emotional, cognitive, and physical well-being. Recent advances in machine learning (ML) and deep learning (DL) have enabled automated and objective approaches to detect and classify depression across multiple modalities, including facial expressions, speech, EEG signals, and textual data. This paper presents a comprehensive review of recent ML-based depression detection techniques, focusing on model architectures such as CNNs, LSTMs, hybrid models, EfficientNet, Vision Transformers, VGG16, and ResNet50. Comparative analysis of publicly available datasets, including DAIC-WOZ, MODMA, and RAVDESS, reveals that multimodal approaches consistently achieve higher accuracy (up to 99%) than single-modality methods. The study identifies critical gaps, such as the need for real-time detection, explainable AI, lightweight models, and more diverse datasets, and highlights future directions to improve the scalability, robustness, and interpretability of ML-based depression detection systems.