stress detection

Stress is an important physiological and psychological reaction which, when prolonged, has a negative impact on the health and well-being of humans. Traditional stress assessment methods tend to be subjective and are not scalable in real-time. In this research, a lightweight machine learning framework to predict stress from easily measurable environmental and activity-related features, temperature, humidity, and steps, is introduced. The proposed method is different from the approach that depends on complex physiological signals and uses minimal non-invasive inputs that can be applied in the wearable and IoT field. The stress level is considered as the target variable, and all the features used as input variables are independent from each other to prevent data leakage. There are several machine learning and deep learning models and their performances are measured, such as LightGBM, Support vector machine (SVM), AdaBoost, Gated Recurrent Unit (GRU). Bayesian hyperparameter optimization is used to improve the performance of the model and guarantee strong generalization. Experimental results show that LightGBM outperforms the other models, giving an accuracy of 97% and a balanced precision, recall and F1 score of 95%. The results demonstrate promising predictive performance on the evaluated dataset. However, since the model was validated on a relatively small publicly available dataset, further evaluation using larger and more diverse real-world datasets is required before practical deployment.

Stress is a major contributor to physical and mental health disorders, underscoring the need for reliable and interpretable detection systems. We present an interpretable multimodal deep learning framework that integrates electrocardiogram (ECG), electrodermal activity (EDA), and respiration signals using modality-specific neural architectures complemented by feature-based classical models. To ensure robustness and generalizability, the framework was evaluated under both subject-independent and subject-dependent protocols. Results show that multimodal fusion consistently outperformed single modalities, achieving 94.8% accuracy, 93.9% F1-score, and 0.97 AUC under subject-independent validation, compared with 86–91% accuracy for individual signals. In subject-dependent experiments, performance reached 97.2% accuracy and 96.8% F1-score. Comparative analysis with recent state-of-the-art methods demonstrated the competitiveness of our approach, while ablation studies confirmed the contribution of each modality and fusion strategy. Statistical significance testing with ( p < 0.05) further validated the reliability of these gains. Importantly, interpretability analyses using Grad-CAM and attention visualization provided physiological insights, revealing that the model emphasizes clinically meaningful regions such as ECG R-peaks and phasic EDA fluctuations. By uniting high performance, robustness, and interpretability, this work advances stress detection research and offers a physiologically grounded framework for continuous mental health monitoring.

Stress is a highly individualized and dynamic psychophysiological response, making reliable continuous monitoring challenging.Existing artificial intelligence approaches to stress detection have demonstrated the usefulness of physiological and neural signals, including electrodermal activity (EDA), heart rate variability (HRV), and electroencephalography (EEG).However, many existing approaches rely on population-level models or limited personalization, while longitudinal validation and adaptive modeling remain comparatively underexplored.This research proposes an adaptive, personalized AI framework for longitudinal stress monitoring by integrating complementary neural and physiological signals from EEG, HRV, and EDA.The proposed framework will initially establish an individual's baseline characteristics and subsequently identify deviations from their personal baseline rather than relying solely on generalized stress patterns across individuals.Machine learning techniques will be investigated for multimodal feature fusion and adaptive model updating as additional individual data becomes available.The study will compare the performance of conventional population-level models with personalized and progressively adaptive models using appropriate evaluation metrics.The research aims to determine whether incorporating individual baseline characteristics and continuous adaptation can improve the reliability and generalizability of stress monitoring.The proposed framework is intended as a foundation for future noninvasive, personalized, and wearable AI-based stressmonitoring systems, while addressing current challenges associated with inter-individual variability, multimodal integration, and longitudinal assessment.

استرس یک پاسخ روان‌فیزیولوژیک بسیار فردی و پویا است که پایش پیوسته و پایا را دشوار می‌سازد. رویکردهای هوش مصنوعی موجود در شناسایی استرس، سودمندی سیگنال‌های فیزیولوژیک و عصبی از جمله فعالیت الکترودرمال (EDA)، تغییرپذیری ضربان قلب (HRV) و الکتروانسفالوگرافی (EEG) را نشان داده‌اند. با این حال، بسیاری از رویکردهای موجود بر مدل‌های مبتنی بر جمعیت یا شخصی‌سازی محدود تکیه دارند، در حالی که اعتبارسنجی طولی و مدل‌سازی تطبیقی نسبتاً کمتر مورد بررسی قرار گرفته‌اند. این پژوهش یک چارچوب هوش مصنوعی تطبیقی و شخصی‌ساز برای پایش طولی استرس با ادغام سیگنال‌های عصبی و فیزیولوژیک مکمل از EEG، HRV و EDA پیشنهاد می‌کند. چارچوب پیشنهادی در ابتدا ویژگی‌های پایه هر فرد را تعیین کرده و سپس انحرافات از خط پایه شخصی را شناسایی می‌کند، به جای آنکه صرفاً بر الگوهای استرس تعمیم‌یافته میان افراد تکیه کند. تکنیک‌های یادگیری ماشین برای همجوشی ویژگی‌های چندوجهی و به‌روزرسانی تطبیقی مدل با در دسترس قرار گرفتن داده‌های فردی بیشتر مورد بررسی قرار خواهند گرفت. این مطالعه عملکرد مدل‌های متداول مبتنی بر جمعیت را با مدل‌های شخصی‌ساز و تدریجاً تطبیقی با استفاده از معیارهای ارزیابی مناسب مقایسه خواهد کرد. هدف پژوهش تعیین این است که آیا گنجاندن ویژگی‌های خط پایه فردی و تطبیق پیوسته می‌تواند پایایی و تعمیم‌پذیری پایش استرس را بهبود بخشد یا خیر. چارچوب پیشنهادی به عنوان بنیانی برای سامانه‌های آینده پایش استرس مبتنی بر هوش مصنوعی، غیرتهاجمی، شخصی‌ساز و پوشیدنی در نظر گرفته شده است، در حالی که به چالش‌های فعلی مرتبط با تفاوت‌های بین‌فردی، ادغام چندوجهی و ارزیابی طولی می‌پردازد.