A Comparative Review University Students Mental Stress Detection Using Machine Learning

Abstract

Depression is a prevalent mental health disorder affecting millions globally. Early detection is crucial for effective intervention and treatment. This paper presents a comprehensive system for detecting depression levels by integrating multimodal data sources, including text, audio, and visual inputs. Utilizing advanced deep learning techniques and natural language processing, the system aims to provide accurate and timely assessments of an individual’s mental health status. The proposed system leverages convolutional neural networks (CNNs) for facial emotion recognition, recurrent neural networks (RNNs) for audio analysis, and transformer-based models for textual data processing. By combining these modalities, the system achieves a holistic understanding of the user’s emotional state, enhancing the accuracy of depression detection. Experimental results demonstrate the efficacy of the system in real-world scenarios, highlighting its potential as a tool for mental health professionals and individuals alike

Introduction

1. Introduction

Depression is a major public health issue, traditionally diagnosed via subjective interviews and questionnaires. Recent advances in machine learning (ML) and deep learning (DL) now enable automated, objective systems for early depression detection by analyzing text, audio, and visual data.

2. Literature Review

Numerous studies show the value of AI in depression detection:

• Multimodal systems using BERT (text), VGG-16 (audio), and ResNet-50 (visual) improve diagnostic accuracy.

• Large Language Models (LLMs) are effective in analyzing transcripts using tools like PHQ-8.

• Social media analysis with sentiment analysis and word embeddings accurately identifies depressive tendencies.

• The DORIS system integrates psychological criteria (DSM-5) with ML for superior depression detection.

3. Motivation

Depression is widespread and underdiagnosed due to limitations in traditional methods (e.g., subjectivity, time, and accessibility). AI can enable:

• Faster, scalable, and objective screening

• Integration of multiple cues (voice, text, facial expression)

• Early intervention and improved outcomes

4. Methodology

The proposed system uses a multimodal deep learning framework:

• Text Analysis: Uses BERT for language modeling and sentiment detection.

• Audio Analysis: Applies CNNs/RNNs to detect prosodic features (pitch, tone).

• Visual Analysis: Employs CNNs and FACS for facial emotion recognition.

• Fusion: Combines all features into a comprehensive vector for final classification via neural networks.

5. Proposed System Design

System components include:

1. Data Collection: Text, audio, and video input from users.

2. Preprocessing: Cleansing, normalization, noise reduction, landmark alignment.

3. Feature Extraction: Deep learning-based processing per modality.

4. Fusion Module: Integrates features using attention or concatenation.

5. Classifier: Predicts depression severity.

6. User Interface: Enables interaction and displays feedback.

6. Results

The system was tested on a multimodal dataset and outperformed unimodal systems, showing better performance across accuracy, precision, recall, and F1-score.

7. Objectives

• Create a robust, automated depression detection system.

• Use advanced deep learning across modalities.

• Improve detection accuracy and reliability.

• Design a user-friendly interface.

8. Applications

• Clinical mental health screening

• Monitoring patients with depression history

• Telemedicine integration

• Student mental health tracking in schools/universities

Conclusion

This paper presents a comprehensive system for detecting depression levels by integrating text, audio, and visual data using advanced deep learning techniques. The multimodal approach enhances the accuracy and reliability of depression detection, offering a valuable tool for mental health professionals and individuals. Future work will focus on expanding the dataset, improving model robustness, and integrating additional modalities such as physiological signals.

References

[1] Makiuchi, Y., et al. (2024). Harnessing multimodal approaches for depression detection using the EDAIC dataset. Nature Mental Health, 3(2), 112. [2] Ray, S., et al. (2024). Using Large Language Models to Detect Depression From User-Generated Diary Text. Journal of Medical Internet Research, 26(1) , e54617. [3] Screening for Depression Using Natural Language Processing. (2024). International Journal of Medical Research, 12(1), e55067. [4] Depression Detection from Social Media Text Analysis using Natural Language Processing. (2024). ACM Digital Library. [5] Lee, Y. S., et al. (2024). DORIS: Detection of Depression via Emotional Text Analysis. IEEE Transactions on Affective Computing.

Copyright

Copyright © 2025 Prof. Rahul M. Dhokane, Viki Punjaram Shirsath, Vinay Babasaheb Pansare, Durgesh Deepak Shinde, Jivan Suresh Patil . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.