Fusion of RNN and LSTM for the Identification of Mental Stress in EEG Signal

Abstract

Proficiently managing stress and recognising it promptly are crucial for facilitating recuperation and averting additional difficulties. Traditional techniques for evaluating anxiety—such as interviews, stress-related enquiries to understand an individual\'s mental condition, and the analysis of facial expressions including eyebrow movement, pupil dilation, or rapid eye blinking—exhibit limitations and may not encompass all manifestations of stress. The electroencephalogram (EEG), a contemporary physiological method, demonstrates potential as an effective instrument for detecting stress in daily environments, attributed to the increasing availability and cost-effectiveness of commercial EEG headsets. This research utilised machine learning methodologies to categorise stress levels derived from resting-state EEG data. Data was obtained from the MathWorks® EEGLAB toolkit and a custom dataset of 20 participants, assembled through surveys and measurements from the Neurosky Mindwave EEG headset. Stress classification was conducted utilising support vector machines (SVM), recurrent neural networks (RNN), long short-term memory (LSTM) networks, and an innovative hybrid approach integrating RNN and LSTM into a parallel fusion model. MATLAB simulation results demonstrate that the suggested method is superior in speed and accuracy compared to other machine learning models, with a 95% accuracy rate—reflecting an improvement of up to 15% over prior methodologies.

Introduction

The text discusses mental stress detection using EEG signals and machine learning, focusing on why early identification of stress is important and how brain signal analysis can support it.

It begins by explaining that mental stress has become a major global health issue, especially after the COVID-19 pandemic, and can lead to serious physical and psychological problems if not detected early. Traditional medical techniques like PET, MRI, ECG, and EMG are used to measure stress, but EEG (Electroencephalography) is highlighted as the most suitable method because it is non-invasive, low-cost, fast, and provides high temporal resolution of brain activity.

Stress is described as a biological “fight-or-flight” response that affects both mental and physical health, leading to symptoms such as anxiety, headaches, hypertension, mood swings, and behavioral changes. Long-term stress can also contribute to chronic diseases like cardiovascular and gastrointestinal disorders, making early detection essential.

The study proposes detecting stress using EEG signals analyzed with a machine learning approach based on a parallel fusion RNN–LSTM model, and compares it with traditional models such as SVM, RNN, and LSTM. The system is implemented using the MATLAB EEGLAB Toolbox, and EEG data is collected using the Neurosky MindWave EEG headset on a small dataset.

The EEG section explains brain wave types and their frequency ranges:

• Delta (0.5–4 Hz): deep sleep
• Theta (4–7 Hz): drowsiness and early sleep
• Alpha (8–12 Hz): relaxed wakefulness
• Beta (13–30 Hz): active thinking and alertness
• Gamma and high-frequency oscillations (30–500 Hz): complex cognitive activity and neurological indicators

These signals reflect different mental states and are used to detect stress-related changes in brain activity.

The methodology uses EEG signal processing, feature extraction, and classification through deep learning models. The proposed RNN–LSTM fusion model aims to improve accuracy by capturing both temporal dependencies and sequential patterns in brain signals.

Overall, the study emphasizes that EEG-based deep learning systems offer a promising, non-invasive, and efficient approach for early mental stress detection, with potential applications in healthcare, neuroscience, and real-time mental health monitoring.

Conclusion

The Smart Safety Driver Device presents a practical and effective approach for minimizing road accidents associated with driver fatigue, alcohol influence, and vehicle fire hazards by combining multiple safety mechanisms within a single embedded system. The microcontroller-based platform continuously acquires and evaluates data from the eye-blink sensor, alcohol sensing module, and flame detector to identify unsafe conditions in real time. Upon detection, the system not only provides immediate alerts through an audible buzzer but also initiates preventive control actions such as restricting ignition during alcohol detection and activating a water-based suppression mechanism in case of fire. Experimental evaluation demonstrates that the system operates with quick response and consistent reliability, thereby enhancing driver awareness and overall vehicle safety. In comparison with conventional systems that focus on individual safety parameters, the proposed integrated design offers improved functionality, reduced system complexity, and cost-effective implementation suitable for a wide range of applications, including public transportation, logistics operations, and private vehicles. Furthermore, the system can be extended with advanced features such as IoT-based connectivity and GPS-enabled emergency notifications, enabling the development of a more intelligent and scalable vehicle safety solution.

References

[1] Krishnamoorthy, Ramesh, Kalimuthu Krishnan, and Srinath Balasubramanian. \"Driver Assistance and Safety System for Accident Prevention Using Embedded Automotive Sensors Integration.\" Ilkogretim Online 20.1 (2021). [2] Nair, Akhil, et al. \"A review on recent driver safety systems and its emerging solutions.\" International Journal of Computers and Applications 46.3 (2024): 137-151. [3] Vinod, Yegireddi Satya, Thoram Saran Kumar, K. Baboji, V. Venkata Lakshmi Dadala, K. Kalyani, Venkata Ramana Kammampati, and U. S. B. K. Mahalaxmi. \"Performance Analysis of Tera Hertz Frequencies on Intelligent Reflecting Surfaces for 6G Communications.\" [4] Kumar, Thoram Saran, Durga Prasad Siddani, I. Rama Satya Nageswara Rao, P. Harika, Anuragh Vijjapu, G. Prasanna Kumar, and B. V. V. Satyanarayana. \"Low power adder circuit design and implementation using LECTOR technique.\" In International Conference on Cognitive Computing and Cyber Physical Systems, pp. 159-170. Cham: Springer Nature Switzerland, 2024. [5] Prasanna Kumar, G., Prudhvi Raj Budumuru, A. Krishna Chaitanya Varma, B. V. V. Satyanarayana, Thoram Saran Kumar, and Inukonda Rama Satya Nageswara Rao. \"Design Analysis of Memristor-Based 7T SRAM Using Heterojunction Tunneling Transistors.\" Journal of Circuits, Systems and Computers 34, no. 17 (2025): 2550280. [6] Lee, J., & Chien, S. ( 2020 ). Driver Drowsiness Detection Systems: A Review. International Journal of Intelligent Transportation Systems Research, 18 ( 2 ), 98 - 115. [7] Shin, J., & Kim, Y. ( 2021 ). Advanced Driver Assistance Systems for Road Safety. IEEE Transactions on Intelligent Transportation Systems, 22 ( 6 ), 3595 - 3604. [8] Bhattacharyya, D., & Sarkar, R. ( 2020 ). Alcohol Detection Systems for Vehicle Security. International Journal of Engineering & Technology, 9 ( 1 ), 319 - 324. [9] Elakkiya, R., & Sathya, S. ( 2019 ). Driver Drowsiness Detection Using EEG Signals and Machine Learning Algorithms. Computers, Materials & Continua, 60 ( 2 ), 697 - 709. [10] Banerjee, R., & Dey, A. (2020). Monitoring Driver Distraction with IoT-enabled Systems for Safe Driving. Journal of Advanced Transportation, 2020, 1 - 11

Copyright

Copyright © 2026 Jitendra Kumar, Dr. Sweety Maniar . 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.