Evaluating different configurations of machine learning models and their transfer learning capabilities for stress detection using heart rate

Journal of Ambient Intelligence and Humanized Computing, Journal Year: 2022, Volume and Issue: 14(8), P. 11011 - 11021

Published: Aug. 27, 2022

Abstract In the twentyfirst-century society, several soft skills are fundamental, such as stress management, which is considered one of key ones due to its strong relationship with health and well-being. However, this skill hard measure master without external support. This paper tackles detection through artificial intelligence (AI) models heart rate, analyzing in WESAD SWELL-KW datasets five supervised unsupervised anomaly that had not been tested before for detection. Also, we analyzed transfer learning capabilities AI since it an open issue field. The highest performance on test data were Local Outlier Factor (LOF) F1-scores 88.89% 77.17% SWELL-KW, Multi-layer Perceptron (MLP) 99.03% 82.75% SWELL-KW. when evaluating these models, MLP performed much worse other dataset, decreasing F1-score 28.41% 57.28% WESAD. contrast, LOF reported better achieving 70.66% 85.00% Finally, found training both (i.e., from different contexts) improved average their generalization; setup, achieved 87.92% 85.51% WESAD, 78.03% 82.16% SWELL-KW; whereas obtained 78.36% 81.33% 79.37% 80.68% Therefore, suggest a promising direction use or multi-contextual improve field, novelty literature. We believe combined non-invasive wearables can enable new generation management mobile applications.

Language: Английский

---

Wearable Biosensor Technology in Education: A Systematic Review

Sensors, Journal Year: 2024, Volume and Issue: 24(8), P. 2437 - 2437

Published: April 11, 2024

Wearable Biosensor Technology (WBT) has emerged as a transformative tool in the educational system over past decade. This systematic review encompasses comprehensive analysis of WBT utilization settings 10-year span (2012–2022), highlighting evolution this field to address challenges education by integrating technology solve specific challenges, such enhancing student engagement, monitoring stress and cognitive load, improving learning experiences, providing real-time feedback for both students educators. By exploring these aspects, sheds light on potential implications future learning. A rigorous search major academic databases, including Google Scholar Scopus, was conducted accordance with PRISMA guidelines. Relevant studies were selected based predefined inclusion exclusion criteria. The articles assessed methodological quality bias using established tools. process data extraction synthesis followed structured framework. Key findings include shift from theoretical exploration practical implementation, EEG being predominant measurement, aiming explore mental states, physiological constructs, teaching effectiveness. biosensors are significantly impacting field, serving an important resource educators students. Their application transform optimize practices through sensors that capture biometric data, enabling implementation metrics models understand development performance professors environment, well gain insights into process.

Language: Английский

---

Wearable Biosensor Technology in Education: A Systematic Review

Published: March 14, 2024

Wearable Biosensor Technology (WBT) has emerged as a transformative tool in the educational system over past decade. This systematic review encompasses comprehensive analysis of WBT utilization settings 10-year span (2012-2022), highlighting both evolution this field and its integration to address challenges education by integrating technology solve specific challenges, such enhancing student engagement, monitoring stress cognitive load, improving learning experiences, providing real-time feedback for students educators. By exploring these aspects, sheds light on potential implications future learning. A rigorous search major academic databases, including Google Scholar Scopus, was conducted accordance with PRISMA guidelines. Relevant studies were selected based predefined inclusion exclusion criteria. The articles assessed methodological quality bias using established tools. process data extraction synthesis followed structured framework. Key findings include shift from theoretical exploration practical implementation EEG being predominant measurement, aiming explore mental states, physiological constructs, teaching effectiveness. biosensors are significantly impacting field, serving an important resource educators students. Their application transform optimize practices through sensors that capture biometric data, enabling metrics models understand development performance professors environment, well gain insights into process.

Language: Английский

---

Academic stress detection based on multisource data: a systematic review from 2012 to 2024

Interactive Learning Environments, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 27

Published: Aug. 6, 2024

The field of academic stress detection has gained significant attention recently because mental and physical health is crucial for success. goal to identify a student's level during the learning process using observable markers including physiological, behavioral, psychological data. In recent years, methods that utilize wearable nonwearable sensors have increased owing their rich functionalities. order discover contemporary developments, coping strategies, limitations, difficulties, potential research areas addressing in educational settings, this study conducted an exhaustive review existing literature. First, we discussed how stressful events influence students' as well statistics frequently utilized monitor stress. Then, machine deep methods, described models. addition, self-regulated strategy, computer-supported strategy interactive technology-supported strategy. This comprehensive analysis latest techniques recommendations avenues tackling settings will help other researchers carry out assess user build systems.

Language: Английский

---

Stress Detection from Facial Expressions Using Transfer Learning Techniques

Published: March 15, 2024

In today's fast paced world stress has become a significant concern, impacting the individual mental health and overall well being. Various models are designed for detecting from facial expressions one among them is ResNet-101 architecture that to detect in real-time video surveillance using symptoms of cues with an accuracy 80.4%. The limitation model minute motions not detected. To overcome these challenges comprehensive evaluation made, evaluating capacity deep learning architectures capturing associated stress. Transfer proven technique which reuses weights pre-trained enhancing capabilities. this research project, we propose development detection system Mini Xception, VGG-16 models. Following number tests, it was shown VGG16 performed best at recognizing tense emotions when combined convolutional layer-based classifier 97.5%.

Language: Английский

---

Detection and monitoring of stress using wearables: a systematic review

Frontiers in Computer Science, Journal Year: 2024, Volume and Issue: 6

Published: Dec. 18, 2024

Over the last few years, wearable devices have witnessed immense changes in terms of sensing capabilities. Wearable devices, with their ever-increasing number sensors, been instrumental monitoring human activities, health-related indicators, and overall wellness. One area that has rapidly adopted is mental health well-being area, which covers problems such as psychological distress. The continuous capability allows detection stress, thus enabling early problems. In this paper, we present a systematic review different types sensors used by researchers to detect monitor stress individuals. We identify detail tasks data collection, pre-processing, features computation, training model explored research works. each step involved monitoring. also discuss scope opportunities for further deals management once it detected.

Language: Английский

---

Time Series Adaptation Network for Sensor-Based Cross Domain Human Activity Recognition

2022 International Joint Conference on Neural Networks (IJCNN), Journal Year: 2023, Volume and Issue: unknown, P. 1 - 8

Published: June 18, 2023

Domain adaptation can apply knowledge learned from the source domain to target by reducing data distribution discrepancy inter domains. However, existing algorithms do not as well on sensor datasets image because of neglect intra discrepancy. The long time collecting a raw segment sensors will lead shift with time, and change variety wearing positions, causing series To solve this problem, we design new model, Time Series Adaptation Network (TSAN), loss, Contrastive Loss (TCL). TSAN uses siamese network "pack" samples divided same into network. Furthermore, TCL is defined similarity "unpack" output, which leads model learn time-independent features. In particular, be used plug-in combine algorithms, so discrepancies considered simultaneously. We conduct extensive experiments eight sensor-based cross human activity recognition (HAR) tasks, including three Routine Activity Recognition (RAR) four Parkinson's tremor Detection (PD) datasets. results show that all are improved an average 5.4% (RAR), 2.2% TSAN.

Language: Английский

---