Deep learning-based stress detection using multimodal biosignals

Q: What are the key findings of this study?

An LSTM-based deep learning model using multimodal biosignals achieved 94.14% accuracy in detecting acute stress, significantly outperforming a DNN-based model (86.42%, p<0.01).

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Acute stress (n=48). Intervention: LSTM-based deep learning model vs. DNN-based model. Primary outcome: Stress detection accuracy (p=<0.01).

Q: What does this research mean for the field?

An LSTM-based deep learning model utilizing multimodal biosignals can detect psychological stress with 94.14% accuracy, outperforming static DNN models, and accurately reflects stress variations associated with gender, smoking, and academic factors. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Q: What question did this study set out to answer?

To develop and validate a real-time deep learning stress detection system combining synchronized multimodal physiological signals and survey data.

Büşra Kocaçınar; Çağla Çöpürkaya; Elif Meriç; Elif Berra Erik; Fatma Patlar Akbulut; Cagatay Catal

doi:10.1007/s41060-026-01270-y

September 2, 2026International Journal of Data Science and AnalyticsOpen Access

Deep learning-based stress detection using multimodal biosignals

View Full Paper

Ask AI

Bookmark

Share

Key result

An LSTM model using multimodal biosignals achieves ~94% accuracy in detecting acute stress, outperforming DNNs.

P<0.01
n=48

Why the study?

Stress-related health problems drive an urgent need for sensitive, real-time monitoring devices capable of capturing psychophysiological stress dynamics in natural settings.

Key Points

To develop and validate a real-time deep learning stress detection system combining synchronized multimodal physiological signals and survey data.
Monitored university students undergoing induced stress via controlled psychological sessions and academic examinations.
Evaluated two model architectures: a deep neural network (DNN) for static physiological patterns and a long short-term memory (LSTM) network for temporal biosignal dependencies.
The LSTM-based model outperformed the standard DNN model, achieving 94.14% classification accuracy versus 86.42% accuracy.
Statistically significant stress score elevations were identified for males compared to females (males: μ = 2.7 vs. females: μ = 2.2, p < 0.05) and smokers compared to nonsmokers (smokers: μ = 3.7 vs. nonsmokers: μ = 2.2, p < 0.05).

Study Design

Type

Observational (n=48)

Multicenter

No

Structured PICO

P

Population

48 undergraduate students (mean age 22.3 years, 41.7% female) who underwent controlled psychological stress induction and real examination environments to evaluate biosignal-based stress detection.

I

Intervention

Long short-term memory (LSTM)-based deep learning model using synchronized physiological signals and survey data

C

Comparator

Deep neural network (DNN)-based model

O

Outcome

Stress detection accuracy

Main Result

Absolute Event Rate: 94.14% vs 86.42%

p-value: p=<0.01

An LSTM-based deep learning model using multimodal biosignals accurately detects stress, outperforming a static DNN model.

Limitations

Small sample size, particularly for the real in-class examination condition (n=8).
Use of synthetic data (SMOTE) for class balancing may not fully capture the complexity and variability of real stress responses.
Risk of overfitting with more expressive deep learning models.
Need for further validation on larger, more diverse populations for real-world generalization.

Implication

LSTM-based stress detection supports temporal biosignal modeling; leaves open prospective validation before clinical or cardiovascular use.

Abstract

Stress-related health problems drive an urgent need for sensitive, real-time monitoring devices capable of capturing psychophysiological stress dynamics in natural settings. This study presents a stress detection system that uses deep learning, combining synchronized physiological signals and survey data from university students who experienced controlled stress through psychological sessions and examinations. We analyzed two model families with differing complexities: a deep neural network (DNN)-based model for static physiological patterns and a long short-term memory (LSTM)-based model designed to capture temporal dependencies in biosignals. The DNN (86.42% accuracy) was outperformed by the LSTM-based model (94.14% accuracy). The statistical results showed a relationship between stress and gender (females: μ μ = 2.2, males: μ μ = 2.7, p < 0.05) and smoking (smokers: μ μ = 3.7, nonsmokers: μ μ = 2.2, p < 0.05). Academic elements also impacted stress perception. The stress scores were in accordance with self-reported measurements, which validated the model.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Kocaçınar et al. (2026) conducted an observational in Acute stress (n=48). LSTM-based deep learning model vs. DNN-based model was evaluated on Stress detection accuracy (p=<0.01). An LSTM-based deep learning model using multimodal biosignals achieved 94.14% accuracy in detecting acute stress, significantly outperforming a DNN-based model (86.42%, p<0.01).

synapsesocial.com/papers/6a98e48e15797dc76eaa8548 https://doi.org/10.1007/s41060-026-01270-y

View Full Paper

Ask AI

Bookmark

Share

Deep learning-based stress detection using multimodal biosignals

Key Points

Study Design

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Deep learning-based stress detection using multimodal biosignals

Key Points

Study Design

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading