The widespread use of social media platforms has resulted in the continuous generation of large-scale textual data reflecting users’ emotions, behaviors, and psychological states. This has created new opportunities for detecting mental health conditions such as depression, anxiety, and stress using machine learning techniques. Based on a comprehensive review and integration of 25 research papers, this study proposes a hybrid machine learning framework for mental health detection from social media data. The approach combines Natural Language Processing (NLP) techniques for data preprocessing and feature extraction with advanced deep learning models, particularly Convolutional Neural Networks (CNN) and Long Short- Term Memory (LSTM) networks. The methodology leverages NLP techniques such as tokenization, stop-word removal, and stemming to transform unstructured textual data into meaningful representations. CNN is employed to extract significant textual features, while LSTM captures temporal dependencies and emotional progression within user- generated content. Comparative analysis of existing models, including Naïve Bayes, Support Vector Machines, CNN, and LSTM, indicates that hybrid architectures consistently outperform standalone models in terms of accuracy and robustness. Experimental results, supported by performance graphs and model evaluations, Demonstrate that the proposed CNN-LSTM model achieves approximately 93% accuracy, showing a significant improvement over traditional machine learning approach. The results are justified through detailed analysis of feature extraction efficiency, sequence modelling capability, and reduced Misclassification rates. This research highlights the effectiveness of hybrid deep learning dynamic emotional patterns in social media. The proposed system is scalable and can be applied to real-time mental health monitoring and early intervention systems.
v et al. (2026) studied this question.