Psychological Stress Classification Using EEG and ECG: A CNN Based Multimodal Fusion Model

Psychological Stress Classification Using EEG and ECG: A CNN Based Multimodal Fusion Model | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Psychological Stress Classification Using EEG and ECG: A CNN Based Multimodal Fusion Model Ben Zhou, Lei Wang, Chenyu Jiang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4015916/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Psychological stress cannot be ignored in today's society, and there is an urgent need for an objective and cost-effective method to detect it. However, traditional machine learning methods that require manual feature extraction require a lot of research time and cannot guarantee accuracy. In this paper, we establish a four-category stress multimodal dataset by collecting EEG and ECG signals from 24 subjects performing mental arithmetic tasks with different difficulty levels and propose a multimodal decision fusion model based on Convolution Neural Network to classify the data. The prediction probabilities of EEG and ECG signals for the four stress categories are first extracted by two models each and then fused into the decision model for the final classification, 5-fold cross-validation and Leave-Three-Subjects-Out experiments are performed, which achieve 91.14% and 91.97% accuracy, respectively. In addition, the features of the convolution layer were visualized using the 1D-Grad-CAM method to improve the interpretability of the model. Multimodel Fusion Convolution Neural Network(CNN) Stress Classification Electroencephalogram(EEG) Electrocardiogram(ECG) Decision Fusion 1D-Grad-CAM Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1. Introduction Stress is a physiological or psychological response to internal or external stressors and a normal reaction to everyday pressure. For individuals, moderate stress can be advantageous for work and study, but excessive stress could lead to a range of illnesses, including cardiovascular, respiratory, and psychological diseases[1-2]. Early intervention during the development of severe stress can markedly improve physical and mental health as well as social stability. Therefore, the development of a stress detection method is essential for effective intervention. Traditionally, researchers measure stress by designing scales that evaluate subjects' perceptual, behavioral, and physical responses. However, accurately gauging an individual's stress level with just one or a few scales can prove challenging, as stress is highly personalized, and the scales' practical usefulness depends on the individual's subjective level of engagement. This limitation leads to the drawbacks of the conventional approach. Objective tests using biomarkers or signals, such as salivary cortisol[3], electroencephalogram(EEG), electrocardiogram(ECG), and others, provide more compelling evidence. Among them, EEG and ECG are favored by researchers due to their non-invasiveness, accessibility, low cost, and high temporal resolution, these signals provide real-time information on how the human body handles stress. Perez-Valero et al.[4] extracted the power density spectra of EEG signals to categorize stress and explored the effect of smoothing the power density spectra on the classification performance. Jebelli et al.[5] developed an online multitask learning algorithm to analyze stress in near real-time by using EEG features in time and frequency domains as well as frontal EEG asymmetry. Wen et al.[6] extracted the power densities of the Theta, Alpha, and Beta bands of EEG to classify the stress and suggested that the absolute power of the Beta band of the right frontal lobe is an effective feature for studying the stress. In traditional machine learning models for ECG-based stress classification, the features related to heart rate variability (HRV) are widely used in various studies[7-9]. In addition to traditional machine learning models, deep learning based EEG and ECG stress classification is also becoming popular. Mane et al.[10] designed a hybrid network of Convolution Neural Network(CNN) and Recurrent Neural Network(RNN) based on EEG signals to binary classify heart stress. Alruily et al.[11] used CNN and Long Short-Term Memory(LSTM) to extract features and combined with the Artificial Bee Colony (ABC) method and African buffalo optimization to classify acute stress and chronic stress based on EEG. Wang et al.[12] designed a convolutional network based on EEG to identify workers' stress in real-time. Giannakakis et al[13] investigated a 1D convolutional network based on ECG to classify psychological stress. Ishaque et al.[14] converted 1D ECG data to 2D images based on migration learning using the CNN network to apply the model to low performance devices to detect psychological stress, which verified the feasibility of the deep learning approach to detect stress in low performance devices. Although the above studies reported acceptable classification performance, they may still not be applicable in practical application scenarios, including factors such as insufficient classification accuracy and insufficient objective data. Multimodal-based studies can effectively address the above issues, such as the combination of ECG and EEG. Attar et al.[15] investigated the relevant features of EEG and ECG by comparing them through statistical analysis and reported features that were statistically significantly different for different levels of stress. Gonzalez-Carabarin et al. [16] investigated the effective features of EEG and ECG signals for stress classification respectively, reporting the most relevant EEG channels for different stress levels and the availability of HRV for stress analysis. Hemakom et al.[17] confirmed that multimodal approaches are more reliable compared to unimodal approaches by extracting features with high EEG and ECG correlations to input into a traditional machine learning model for stress classification. However, these methods rely on manual feature extraction, which may lead to the loss of key features, and it is time-consuming and laborious to try to find effective features, while deep learning multimodal fusion methods based on the stress classification task have rarely been studied. In this study, we gathered data on four levels of psychological stress through the implementation of a mental arithmetic task experiment. We introduce a deep learning model that utilizes EEG and ECG bimodality for the purpose of studying stress categorization. Additionally, we will assess the potential of our multimodal deep learning model by comparing it to conventional machine learning methods. Chapter 2 will detail the data collection procedure and the analysis techniques employed, while Chapter 3 will present the model's efficacy and compare it to other approaches. Finally, the study will conclude with a summary, discussion, and outlook. 2. Materials and Methods 2.1 Data Acquisition Numerous studies have demonstrated that mental arithmetic tasks[18-20] can induce psychological stress, which will be utilized to induce stress in the current experiment. 26 volunteers, aged between 20-30, who were well-educated, physically and mentally healthy, and not abusing any substances, were recruited for the study. Informed consent forms were completed by all participants before the experiment, and the experimental protocol was in accordance with the Declaration of Helsinki. EEG data were collected using a 14-channel wireless EMOTIV EpocX headset with a sampling frequency of 128 HZ. Figure 1 displays the electrode schematic. ECG data were gathered with a single-channel CHERO ECG patch at a sampling frequency of 250 HZ. The software for the experiment was designed using Python3.9 and Pyside6 framework on the Windows 10 platform. The computer controlled the marking signals for the data collection to ensure high temporal precision. The study is composed of five distinct phases: a resting phase, a practice phase, and three formal phases. Firstly, we will gather two minutes of resting state data from each subject, during which they are not required to perform any task but simply remain relaxed. A flow chart depicting the experiment for all phases except the resting phase is included in Figure 2. Each stage comprises n subsections, wherein participants are required to complete m questions within 10 seconds. The questions consist of three two-digit numbers, which are added or subtracted. The computer generates a random result that is close to the correct answer, and subjects are expected to choose whether the result is greater than, less than, or equal to the correct answer using the keyboard. There is a rest period of 5 seconds after each subsection and a 20-second rest period in the middle of each stage. The task difficulty increases with each stage. Ideally, no data indicating stress levels will be collected from the subjects during the resting stage. In stage 1, only light stress data will be collected, while in stage 2, concentrated stress data will be gathered. Heavy stress data will be collected during stage 3. The data obtained from the practicing stage and the resting time will be discarded. During the experiment, participants were instructed to minimize signal artifacts by refraining from making large movements and only applying slight pressure with their fingers on the keyboard. Failure to adhere to these guidelines resulted in the experiment being restarted or terminated, depending on the circumstances. After assessing the data quality of 26 participants, we removed two subjects due to excessive data noise and preserved the data from the remaining 24 participants. For each subject, we selected the central 100 seconds of the resting state data over a 2-minute period and partitioned it into ten equal segments. Additionally, we obtained the data from each subsection of the three phases, yielding a total of 4*10=40 data samples. Each sample included 10 seconds of EEG and ECG data, with 14*1280 sampling points for EEG and 1*2500 sampling points for ECG. We collected a total of 24*40 data samples for 24 subjects, resulting in 960 overall samples. 2.2 Data Pre-processing Since both EpocX and CHERO devices have some denoising of the signal and the denoising method used is not available, this paper will not take any pre-processing measures and will directly use the raw data for analysis to maximize the retention of information. 2.3 Traditional Machine Learning Method 2.3.1 Feature Extraction EEG power band and power asymmetry in brain regions appear to have a strong correlation with psychological stress[4-6, 12]. The current study extracted the frequency band power of Delta (0.5-3.5 Hz), Theta (4-7.5 Hz), Alpha (8-13 Hz), and Beta (14-30 Hz) from 14 EEG channels. Furthermore, the frequency band power of seven symmetrical groups of channels (AF3-AF4, F3-F4, Fc5-FC6, F7-F8, T7-T8, P7-P8, O1-O2, shown in Figure 1) was analyzed for Alpha frequency band power asymmetry, as illustrated in Table 1. Table 1. Description and numbers of extracted EEG features. Totally 63 EEG features in each sample. HRV, which responds to autonomic nervous system activity, has been the primary method for analyzing ECG signals[7-9]. The study presents the time domain, frequency domain, and statistical features extracted in Table 2. Table 2 . Description of extracted ECG features. Totally 12 features in each sample. Feature Description VLF power of very low frequency(<0.04Hz) LF power of low frequency(0.04-0.15Hz) HF power of high frequency(0.15-0.4Hz) LF/HF Ratio of LF and HF RMSSD root mean of squared differences of successive NN Intervals SDNN standard deviation of an NN intervals NN_mean Mean NN interval NN_min Minimum NN interval NN_max Maximum NN interval SDSD standard deviation of differences of successive NN intervals PNN20 NN20 count divided by the total number of all NN intervals Triangular Index triangular index based on the NN intervals histogram 2.3.2 Classification Support Vector Machines (SVMs), Decision Trees (DTs), Logistic Regression (LR), Naive Bayes (NB), K-Nearest Neighbors (KNN), and Stochastic Gradient Descent (SGD) algorithms are utilized for comparing classifications. The classification will be separated into two sessions. Firstly, the EEG and ECG features will be separately introduced to the classification algorithms to demonstrate the unimodal classification results. Subsequently, the multimodal fusion classification results will be presented utilizing feature fusion and decision fusion strategies, respectively. In the feature fusion approach, EEG and ECG features are overlayed and fed into the classifier. In the decision fusion method, both types of features are fed into the same classifier, and the outputs are added to generate the final classification result. Please refer to Figure 3 for the visual representation of these two techniques. 2.4 Deep Learning Method CNN has demonstrated a high level of effectiveness in EEG and ECG signal-related tasks [10-14]. In this study, we utilized models based on CNN to categorize EEG and ECG signals separately, and subsequently incorporated the classification outcomes of both into a decision model for conclusive decision making. Next, we will present EEG, ECG, and categorical decision models in order. 2.4.1 Model of EEG Classification For EEG signals, we used the convolutional model shown in Figure 4: The initial 14-lead EEG signal serves as the model's input and the corresponding data is allotted to the 14 channels of the convolutional layer. Temporal features are then obtained from the Convolution1 Layer, consisting of 28 convolutional kernels with each one measuring (1,3) in size. Next, the first feature dimension (channel) is switched with the last dimension (sampling point). And the feature shape is transformed from (28, 1, 1278) to (1278, 1, 28). Then, the transposed feature is inputted into the Convolution2 Layer, which has 28 convolution kernels with dimensions of (1, 1) to further extract the features of each channel in the Convolution1 Layer. Subsequently, the transposition is reversed to the same dimensions of the feature as the output of Convolution1. Input the feature into the Convolution3 Layer, which contains 56 convolution kernels, each with dimensions of (1,3), to extract deep temporal features. Then, use the channel maximum pooling layer to downscale the feature dimensions from (56,1,1276) to (1,1,1276). Finally, flatten the feature and input it into the Full Connection Layer for classification into four classes. Notably, the convolution process does not involve padding. 2.4.2 Model of ECG Classification Inspired by ST-CNN-GAP-5-Net[21], we designed a convolutional network to analyze ECG signals as shown in Figure 5. The model comprises six convolution modules, and Figure 5 displays the structure of each module in chronological order: Convolution Layer, Batch Normalization Layer, ReLu Layer, and Pooling Layer. The first five modules are Temporal Convolution Modules, while the last one is a Spatial Convolution Module. The kernel size for each filter is (1,5), (1,5), (1,5), (1,5), (1,3), and (1,3), respectively. Additionally, pooling is applied with sizes of (1,2), (1,4), (1,2), (1,4), (1,2), and (1,4). The original ECG signal is processed through various filters, including 4, 8, 16, 32, and 64. The Spatial Convolution Module is then used to extract spatial information from the features. Next, the features undergo downsizing through channel convolution with 64 filters and a kernel size of (12, 1). Lastly, global average pooling is employed. Finally, the extracted features are inputted into the Dense Module, which includes a Full Connection Layer, Batch Normalization Layer, ReLu Layer, and Dropout Layer (dropout rate of 0.1), to generate predicted probabilities for the four classes. Padding is applied throughout the convolution process. 2.4.3 Model of Decision Fusion The predicted probabilities of EEG and ECG for the four categories were acquired from the models mentioned in Sections 2.4.2 and 2.4.3, respectively. Then, all the probable values of the samples in the training set were inserted into the decision fusion model presented in Figure 6 for training. First, the predicted probabilities of EEG and ECG for the four pressure levels are added separately to calculate a new probability. This probability is then combined with the predicted probabilities of EEG and ECG to create a decision matrix with a size of (4,3). The matrix is inputted into the Decision Convolution Model illustrated in Fig. 6. The Convolution1 Layer, with filters of 2 and kernel size of (2,3), will learn the probabilities of EEG, ECG, and their sum to automatically assign weights. Then, the Convolution2 Layer with 4 filters and kernel size (2,1), will extract the lateral decision features again, flatten all features, and input them into the Full Connection Layer to obtain the final classification result. Padding is not used throughout the entire convolution process. 2.5 Performance Metrics Accuracy and Kappa score are commonly used as a measure of model performance and are also considered to be the most important in classification tasks, and will serve as the basis for comparing model performance in this paper. Accuracy is the ratio of correct predictions to total predicted values. To calculate accuracy, it is necessary to first calculate true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) using the following formula: The Kappa coefficient calculates whether the model's prediction accuracy is balanced for each class and is determined by: where Pa is the actual percentage of agreement, and Pe is the expected percentage chance of agreement. 2.6 1D-Grad-CAM Grad-CAM[22] obtains the weight of any convolutional layer on the target category through backpropagation. The original method displays features obtained by the convolutional layer on the input image. This paper combines the weights of the convolutional layer with the input signal and displays information obtained by the convolutional layer in the 1D signal for improved interpretability. For the principle of Grad-CAM derivation, please refer to the original article. 3. Experiments and Results All models were trained and tested on the stressed dataset presented in Chap. 2. The models were compared using a 5-fold cross-validation and Leave Three Subject Out (LTSO) validation method. PyTorch 1.8, Scikit-learn 1.2, and Tensorflow 2.6 were utilized to construct the modeling framework, with NVIDIA GTX 1070 GPU being utilized to expedite the model training. The optimizer used was Adam. 3.1 Leave Three Subjects Out (LTSO) The approach employed is derived from the Leave One Subject Out (LOSO) method. As this experiment only has 40 samples from one subject out of a total of 24 subjects, the amount of data is insufficient for the test set. Therefore, this thesis retains the data of three subjects for the test set, while the data of the remaining 21 subjects is used for the training set. To consider both the experiment's quality and time constraints, we randomly divided the 24 subjects into randomized groups, with three in each group and a total of eight groups. We then used the data from one group as a test set and the other seven as a train set, similar to an 8-fold cross-validation approach. 3.2 Results of Traditional Machine Learning Method In this section, we present the results of classifying EEG and ECG individually and fusing them using conventional machine learning techniques. The 63 previously mentioned EEG features and 12 ECG features undergo normalization using sklearn's StandardScaler and are then utilized as inputs for classification by SVM, KNN, NB, LR, SGD, and DT classifiers. The outcomes of individual classification results will be contrasted with those produced by feature and decision fusion, after which the average accuracy(%) for both the five-fold cross-validation and the LTSO experiments will be displayed in Tables 3 and 4 . Table 3 Comparison of the accuracy of five-fold cross-validation experiments based on traditional machine learning methods, with the best results indicated in bold. Classifier EEG only ECG only Feature fusion Decision fusion SVM 39.06 36.87 43.95 44.89 KNN 45.83 44.79 46.14 55.52 NB 26.45 27.60 30.72 28.02 LR 37.08 34.16 41.56 37.50 SGD 35.21 30.00 34.06 34.06 DT 66.67 53.02 68.43 74.17 Table 4 Comparison of the accuracy of LTSO experiments based on traditional machine learning methods, with the best results indicated in bold. Classifier EEG only ECG only Feature fusion Decision fusion SVM 39.06 37.29 40.11 42.08 KNN 48.95 48.33 51.66 61.25 NB 26.45 30.28 32.08 29.58 LR 38.74 34.16 42.61 38.85 SGD 33.33 26.87 36.66 31.45 DT 71.77 58.54 72.61 82.18 The results reveal that the fusion model yields higher classification accuracy than either EEG or ECG alone in both 5-fold cross-validation and LTSO experiments. This underscores the strengths of multimodal fusion. Furthermore, decision fusion outperforms feature fusion, indicating the former's greater potential. Among the six classifiers, the DT classifier exhibits superior performance, leading with classification accuracies of 74.17% and 82.18% in 5-fold cross-validation and LTSO experiments, respectively. 3.3 Results of The Proposed Deep Learning Method This section first compares the outcomes of the proposed models on an individual basis with those of the leading models in the same field. Afterward, it analyzes the results of the multimodal fusion models as compared to the unimodal ones. 3.3.1 Comparison of EEG Models The proposed EEG model was compared with DeepConvNet[ 23 ], EEGNet[ 24 ], EEGNeX[ 25 ], ATCNet[ 26 ], and the best results (DT) obtained from the previously traditional machine learning model. Table 5 and Table 6 demonstrate the average accuracy and kappa score of the 5-fold cross-validation and LTSO experiments. Table 5 Comparison of classification results from 5-fold cross-validation experiments of the EEG models, with the best results bolded. Model Accuracy(%) Kappa score DT 66.67 0.5547 DeepConvNet 81.54 0.7475 EEGNet 80.94 0.7447 EEGNeX 86.35 0.8176 ATCNet 87.01 0.8322 Proposed EEG model 88.58 0.8592 Table 6 Comparison of classification results from LTSO experiments of the EEG models, with the best results bolded. Model Accuracy(%) Kappa score DT 71.77 0.6236 DeepConvNet 77.29 0.6972 EEGNet 80.62 0.7416 EEGNeX 84.69 0.7886 ATCNet 85.31 0.7992 Proposed EEG model 85.86 0.8111 The findings indicate that deep learning approaches typically outperform traditional machine learning approaches, and the proposed EEG model achieves the best classification performance when used for stress classification. 3.3.2 Results of The Proposed ECG Model Most studies on ECG stress classification focused on analyzing HRV, and only a few have utilized deep learning. Furthermore, few models with open-source code are available. This section presents the ablation study of the proposed model and compares it with the DT method, which was introduced previously. The ablation experiments determine the effects of decreasing or increasing the temporal convolutional module on the classification. The results of the 5-fold cross-validation and LTSO experiments are presented in Tables 7 and 8 , respectively. Table 7 Comparison of classification results from 5-fold cross-validation experiments of ECG model, with the best results bolded, where 4-T-conv-layer and 6-T-conv-layer mean 4 and 6 Temporal Convolution Layer utilized in proposed ECG model. Model Accuracy(%) Kappa score DT 53.02 0.3739 4-T-conv-layer 63.75 0.5158 6-T-conv-layer 82.81 0.7724 Proposed ECG model 85.21 0.8023 Table 8 Comparison of classification results from LTSO experiments of ECG model, with the best results bolded, where 4-T-conv-layer and 6-T-conv-layer mean 4 and 6 Temporal Convolution Layer utilized in proposed ECG model. Model Accuracy(%) Kappa score DT 58.54 0.4472 4-T-conv-layer 66.46 0.5367 6-T-conv-layer 83.97 0.7836 Proposed ECG model 87.29 0.8305 The findings indicated that the ECG deep learning model proposed yields much higher performance than traditional machine learning methods. Additionally, the proposed 5-layer temporal convolution module outperformed both the 4-layer temporal convolution module and the 6-layer convolution module in terms of classification performance. 3.3.3 Comparison of Multimodal and Unimodal Methods Comparison of classification results of multimodal and unimodal decision fusion models are shown in Tables 9 and 10 as follows. Table 9 Comparison of classification results from multimodal and unimodal 5-fold cross-validation experiments, with the best results shown in bold. Model Accuracy(%) Kappa score EEG Only 88.58 0.8592 ECG Only 85.21 0.8023 Proposed Fusion Model 91.14 0.8817 Table 10 Comparison of classification results from multimodal and unimodal LTSO experiments, with the best results shown in bold. Model Accuracy(%) Kappa score EEG Only 85.86 0.8111 ECG Only 87.29 0.8305 Proposed Fusion Model 91.97 0.8931 The results demonstrated that the multimodal fusion model outperforms the unimodal model in terms of classification performance. Additionally, decision fusion enhanced classification performance by automatically assigning weights through the convolutional network, particularly in situations where there was an imbalance in the classification effect of EEG and ECG models. 3.3.4 Feature Visualization In this section, we utilized the 1D-Grad-CAM technique to visualize the proposed convolutional network model and display the features acquired by the convolutional layers which could enhance the interpretability of the model. The second and third convolutional layers of the EEG model will be fed into the 1D Grad-CAM to demonstrate the selected features, and we randomly selected one correctly classified sample from each class of all samples visualized in Fig. 7 : The results indicated that the second convolutional layer of the EEG model prioritizes continuous EEG signal features, while the third convolutional layer was predominantly focused on lower frequency information. This finding is consistent with the majority of studies that manually extract features, including lower frequency band power. The temporal convolutions of the ECG model from the second through the fifth layers, along with the final spatial convolution, were utilized for feature visualization in the 1D Grad-CAM, as illustrated in Fig. 8 . The results indicated that the first three layers of the temporal convolution concentrate on features primarily within the QRS wave range, whereas the fifth layer concentrates on global features, and the spatial convolution focuses on lower frequency data, consistent with the commonly used HRV analysis for ECG. Confirming the model's feature-extraction effectiveness. The last layer of convolutional features of the decision model was also visualized to show the different weights given to the EEG ECG decisions, four correctly identified samples of different categories of stress were randomly selected and the results are shown in Fig. 9 : As can be seen from the figure for no or low stress levels EEG is selected with more weight, while for moderate or severe stress the selection of ECG is more important. 4. Discussion and Conclusion In this study, the performance difference between the traditional machine learning method of manual feature extraction and the convolution-based deep learning method is investigated by collecting EEG and ECG data from 24 subjects with four types of stress levels, which confirms the superior performance of the proposed unimodal model among the existing methods, and the deep learning-based decision fusion model is investigated on the basis of the unimodal model, which further improves the classification performance of the model and confirms the effectiveness of the multimodal model in the pressure detection task. In addition, the features extracted from the convolutional layer are visualized using the Grad-CAM method, which enhances the interpretability of the model. From the results, the experimental LTSO accuracy of ECG is higher than 5-fold cross-validation, and the experimental LTSO accuracy of EEG is lower than 5-fold cross-validation, which means that the intersubject variability of ECG is lower than that of EEG, which may be related to the low signal-to-noise ratio of EEG. Although acceptable classification performance was achieved in this study, the proposed decision fusion model proposed in this study failed to address the degradation of classification performance caused when the difference in unimodal classification performance is too large, for example, at the beginning of the study, a different ECG model was applied for stress classification, but only about 78% accuracy was achieved, which was about 10% different from the accuracy of the EEG model, resulting in the classification accuracy of the decision fusion model failing to exceed that using only EEG. In the future, we will try to overcome this limitation with improved attention mechanisms. Overall, to address the lack of performance in stress classification tasks, deep learning based decision fusion models have great potential. Declarations Ethics approval and consent to participate: Not applicable. Consent for publication : Not applicable. Availability of data and materials: The data that support the findings of this study are available on request from the author, upon reasonable request. Competing interests: The authors declare that they have no competing interests. Funding: This research was funded by Department of Science and Technology of Shandong Province, grant number ZR2020QF018. Author Contributions: Conceptualization, B.Z. and L.W.; methodology, B.Z.; software, B.Z.; validation, L.W. and C.J.; writing—original draft preparation, B.Z.; funding acquisition, L.W. and C.J.; All authors have read and agreed to the published version of the manuscript. Acknowledgement: Not applicable. References Levine, G. N. Psychological Stress and Heart Disease: Fact or Folklore? The American Journal of Medicine 2022, 135 (6), 688–696. https://doi.org/10.1016/j.amjmed.2022.01.053. Hammen, C. Stress and Depression. Annual Review of Clinical Psychology 2005, 1 (1), 293–319. https://doi.org/10.1146/annurev.clinpsy.1.102803.143938. Sinha, R. Chronic Stress, Drug Use, and Vulnerability to Addiction. Ann N Y Acad Sci 2008, 1141, 105–130. https://doi.org/10.1196/annals.1441.030. Perez-Valero, E.; Lopez-Gordo, M. A.; Vaquero-Blasco, M. A. EEG-Based Multi-Level Stress Classification with and without Smoothing Filter. Biomedical Signal Processing and Control 2021, 69, 102881. https://doi.org/10.1016/j.bspc.2021.102881. Jebelli, H.; Mahdi Khalili, M.; Lee, S. A Continuously Updated, Computationally Efficient Stress Recognition Framework Using Electroencephalogram (EEG) by Applying Online Multitask Learning Algorithms (OMTL). IEEE Journal of Biomedical and Health Informatics 2019, 23 (5), 1928–1939. https://doi.org/10.1109/JBHI.2018.2870963. Wen, T. Y.; Mohd Aris, S. A. Hybrid Approach of EEG Stress Level Classification Using K-Means Clustering and Support Vector Machine. IEEE Access 2022, 10, 18370–18379. https://doi.org/10.1109/ACCESS.2022.3148380. Vanitha, L.; Suresh, G. R. Hybrid SVM Classification Technique to Detect Mental Stress in Human Beings Using ECG Signals. In 2013 International Conference on Advanced Computing and Communication Systems; 2013; pp 1–6. https://doi.org/10.1109/ICACCS.2013.6938735. Pourmohammadi, S.; Maleki, A. Stress Detection Using ECG and EMG Signals: A Comprehensive Study. Computer Methods and Programs in Biomedicine 2020, 193, 105482. https://doi.org/10.1016/j.cmpb.2020.105482. Kuttala, R.; Subramanian, R.; Oruganti, V. R. M. Hierarchical Autoencoder Frequency Features for Stress Detection. IEEE Access 2023, 11, 103232–103241. https://doi.org/10.1109/ACCESS.2023.3316365. Mane, S. A. M.; Shinde, A. StressNet: Hybrid Model of LSTM and CNN for Stress Detection from Electroencephalogram Signal (EEG). Results in Control and Optimization 2023, 11, 100231. https://doi.org/10.1016/j.rico.2023.100231. Alruily, M. Sentiment Analysis for Predicting Stress among Workers and Classification Utilizing CNN: Unveiling the Mechanism. Alexandria Engineering Journal 2023, 81, 360–370. https://doi.org/10.1016/j.aej.2023.09.040. Wang, Y.; Huang, Y.; Gu, B.; Cao, S.; Fang, D. Identifying Mental Fatigue of Construction Workers Using EEG and Deep Learning. Automation in Construction 2023, 151, 104887. https://doi.org/10.1016/j.autcon.2023.104887. Giannakakis, G.; Trivizakis, E.; Tsiknakis, M.; Marias, K. A Novel Multi-Kernel 1D Convolutional Neural Network for Stress Recognition from ECG. In 2019 8th International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW); 2019; pp 1–4. https://doi.org/10.1109/ACIIW.2019.8925020. Ishaque, S.; Khan, N.; Krishnan, S. Detecting Stress through 2D ECG Images Using Pretrained Models, Transfer Learning and Model Compression Techniques. Machine Learning with Applications 2022, 10, 100395. https://doi.org/10.1016/j.mlwa.2022.100395. Attar, E. T.; Balasubramanian, V.; Subasi, E.; Kaya, M. Stress Analysis Based on Simultaneous Heart Rate Variability and EEG Monitoring. IEEE Journal of Translational Engineering in Health and Medicine 2021, 9, 1–7. https://doi.org/10.1109/JTEHM.2021.3106803. Gonzalez-Carabarin, L.; Castellanos-Alvarado, E. A.; Castro-Garcia, P.; Garcia-Ramirez, M. A. Machine Learning for Personalised Stress Detection: Inter-Individual Variability of EEG-ECG Markers for Acute-Stress Response. Computer Methods and Programs in Biomedicine 2021, 209, 106314. https://doi.org/10.1016/j.cmpb.2021.106314. Hemakom, A.; Atiwiwat, D.; Israsena, P. ECG and EEG Based Detection and Multilevel Classification of Stress Using Machine Learning for Specified Genders: A Preliminary Study. PLOS ONE 2023, 18 (9), e0291070. https://doi.org/10.1371/journal.pone.0291070. He, J.; Li, K.; Liao, X.; Zhang, P.; Jiang, N. Real-Time Detection of Acute Cognitive Stress Using a Convolutional Neural Network From Electrocardiographic Signal. IEEE Access 2019, 7, 42710–42717. https://doi.org/10.1109/ACCESS.2019.2907076. McDuff, D.; Gontarek, S.; Picard, R. Remote Measurement of Cognitive Stress via Heart Rate Variability. In 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society; 2014; pp 2957–2960. https://doi.org/10.1109/EMBC.2014.6944243. Giannakakis, G.; Grigoriadis, D.; Giannakaki, K.; Simantiraki, O.; Roniotis, A.; Tsiknakis, M. Review on Psychological Stress Detection Using Biosignals. IEEE Transactions on Affective Computing 2022, 13 (1), 440–460. https://doi.org/10.1109/TAFFC.2019.2927337. Anand, A.; Kadian, T.; Shetty, M. K.; Gupta, A. Explainable AI Decision Model for ECG Data of Cardiac Disorders. Biomedical Signal Processing and Control 2022, 75, 103584. https://doi.org/10.1016/j.bspc.2022.103584. Selvaraju, R. R.; Cogswell, M.; Das, A.; Vedantam, R.; Parikh, D.; Batra, D. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization. In 2017 IEEE International Conference on Computer Vision (ICCV); 2017; pp 618–626. https://doi.org/10.1109/ICCV.2017.74. Schirrmeister, R. T.; Springenberg, J. T.; Fiederer, L. D. J.; Glasstetter, M.; Eggensperger, K.; Tangermann, M.; Hutter, F.; Burgard, W.; Ball, T. Deep Learning with Convolutional Neural Networks for EEG Decoding and Visualization. Hum Brain Mapp 2017, 38 (11), 5391–5420. https://doi.org/10.1002/hbm.23730. Lawhern, V. J.; Solon, A. J.; Waytowich, N. R.; Gordon, S. M.; Hung, C. P.; Lance, B. J. EEGNet: A Compact Convolutional Network for EEG-Based Brain-Computer Interfaces. J. Neural Eng. 2018, 15 (5), 056013. https://doi.org/10.1088/1741-2552/aace8c. Chen, X.; Teng, X.; Chen, H.; Pan, Y.; Geyer, P. Toward Reliable Signals Decoding for Electroencephalogram: A Benchmark Study to EEGNeX. Biomedical Signal Processing and Control 2024, 87, 105475. https://doi.org/10.1016/j.bspc.2023.105475. Altaheri, H.; Muhammad, G.; Alsulaiman, M. Physics-Informed Attention Temporal Convolutional Network for EEG-Based Motor Imagery Classification. IEEE Transactions on Industrial Informatics 2023, 19 (2), 2249–2258. https://doi.org/10.1109/TII.2022.3197419. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4015916","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":277763570,"identity":"fd701263-9f39-46d3-900c-9d9a995a3bf6","order_by":0,"name":"Ben Zhou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzElEQVRIie3PMQrCMBTG8RcCcQm4vuLgFSKCOBR6EJdMnXQSREEwLp0EV8VL2Bu8EqhL0LXOXqAHEFRwcWvcBPOf3w/eBxAK/WBtuNSE88dq2zKeJFpTj4aOs/2GPImy1KdFxtmx0r6flaTpmgkO11tewTIeNQq2ISp2ZynYIZ0OoUwnpolwLIzFGUreGQ+QGdtMRPcG9i4Uish5EgklEGZaSZSeBMG9iCON8rVF+2xJyPEa56STk82rehk3k88U6m/O3+RbEQqFQv/REzxDRVEM1oQCAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0009-0007-4004-2490","institution":"Shandong University of Traditional Chinese Medicine","correspondingAuthor":true,"prefix":"","firstName":"Ben","middleName":"","lastName":"Zhou","suffix":""},{"id":277763571,"identity":"7902d083-cd11-4a94-aed8-3c6484122da1","order_by":1,"name":"Lei Wang","email":"","orcid":"","institution":"Suzhou Institute of Biomedical Engineering and Technology","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Wang","suffix":""},{"id":277763572,"identity":"bcb1ea92-6bd5-4dde-9288-099d242ed731","order_by":2,"name":"Chenyu Jiang","email":"","orcid":"","institution":"Suzhou Institute of Biomedical Engineering and Technology","correspondingAuthor":false,"prefix":"","firstName":"Chenyu","middleName":"","lastName":"Jiang","suffix":""}],"badges":[],"createdAt":"2024-03-05 08:20:46","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4015916/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4015916/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":52505184,"identity":"921bbbcb-acd6-4eb0-878a-444e2d1868cb","added_by":"auto","created_at":"2024-03-12 10:57:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":121103,"visible":true,"origin":"","legend":"\u003cp\u003eElectrode location of EMOTIV EpocX headset.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/d1fc9e886586f893cf52cc09.png"},{"id":52505185,"identity":"8a2ce879-210b-441d-9603-2e848af0674c","added_by":"auto","created_at":"2024-03-12 10:57:17","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":117937,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of data acquisition (except resting phase), where n is the number of subsections and m represents the number of topics to be completed in each subsection.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/8d04807de4582aa5d8dca705.png"},{"id":52505191,"identity":"b4adcac3-0f5b-4855-a4ca-ca94429908b4","added_by":"auto","created_at":"2024-03-12 10:57:18","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":101447,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of feature fusion and decision fusion methods.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/bd99c33f4905104b3c48ab64.png"},{"id":52505187,"identity":"29b2b834-b981-43a0-a2f8-885f8310a114","added_by":"auto","created_at":"2024-03-12 10:57:17","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":200351,"visible":true,"origin":"","legend":"\u003cp\u003eModel of EEG Classification, where key parameters for each layer are shown on the boxes they represent.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/b94daf3aab03f040ec2dd575.png"},{"id":52505363,"identity":"0ac00661-db41-44b2-8002-55dc47abc278","added_by":"auto","created_at":"2024-03-12 11:05:17","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":158176,"visible":true,"origin":"","legend":"\u003cp\u003eECG classification model, where every convolutional module has the same structure as shown in the bottom left, with differences in the convolutional kernel size, kernel number, and Pooling Layers at the end.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/c93d222cb688d132dd1f1a24.png"},{"id":52505186,"identity":"90220857-8adf-47a6-a300-e39442f4a0ff","added_by":"auto","created_at":"2024-03-12 10:57:17","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":79926,"visible":true,"origin":"","legend":"\u003cp\u003eDecision fusion model, where \"+\" stands for summation and \"merge\" for superposition, and the main parameters of the convolutional layers are given under the corresponding boxes.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/2a80581e65d2f6e81d6628fd.png"},{"id":52505189,"identity":"87fcd7a5-b397-468b-a049-d8b2f40e8339","added_by":"auto","created_at":"2024-03-12 10:57:17","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":126626,"visible":true,"origin":"","legend":"\u003cp\u003eVisualizing the attention features in the convolutional layers of the proposed model with different EEG stress signals. Darker colors indicate higher attention levels.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/8e411999cfa64b53e57904c7.png"},{"id":52505190,"identity":"3d0cbef6-75f8-4a28-b0db-6b81cc09ad62","added_by":"auto","created_at":"2024-03-12 10:57:17","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":174855,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of the feature concerned by second, third, fourth, and fifth Temporal Convolution Layer and Spatial Convolution Layer of the proposed ECG model. Darker colors indicate higher attention levels.\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/88bbb8e22a3a78fe9e5f9663.png"},{"id":52505192,"identity":"ba58ab91-8ee0-4f1b-935f-8cf72fd4509b","added_by":"auto","created_at":"2024-03-12 10:57:18","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":24848,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of the features obtained from the final convolutional layer of the proposed decision model, with the horizontal coordinates representing the probability of EEG or ECG prediction for different levels of stress, where 1 is No Stress, 2 is Low Stress, 3 is Moderate Stress, and 4 representing Severe Stress. Darker colors indicate higher attention levels.\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/90b6aeeed7945f1cf6e8db83.png"},{"id":55695932,"identity":"bf76f38a-3dd6-403e-8f3e-16dd9fec5209","added_by":"auto","created_at":"2024-05-02 01:27:49","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1921663,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4015916/v1/39ab88d2-a930-4475-a861-f3705cbcf2de.pdf"}],"financialInterests":"","formattedTitle":"Psychological Stress Classification Using EEG and ECG: A CNN Based Multimodal Fusion Model","fulltext":[{"header":"1. Introduction","content":"Stress is a physiological or psychological response to internal or external stressors and a normal reaction to everyday pressure. For individuals, moderate stress can be advantageous for work and study, but excessive stress could lead to a range of illnesses, including cardiovascular, respiratory, and psychological diseases[1-2]. Early intervention during the development of severe stress can markedly improve physical and mental health as well as social stability. Therefore, the development of a stress detection method is essential for effective intervention.\nTraditionally, researchers measure stress by designing scales that evaluate subjects' perceptual, behavioral, and physical responses. However, accurately gauging an individual's stress level with just one or a few scales can prove challenging, as stress is highly personalized, and the scales' practical usefulness depends on the individual's subjective level of engagement. This limitation leads to the drawbacks of the conventional approach. Objective tests using biomarkers or signals, such as salivary cortisol[3], electroencephalogram(EEG), electrocardiogram(ECG), and others, provide more compelling evidence.\nAmong them, EEG and ECG are favored by researchers due to their non-invasiveness, accessibility, low cost, and high temporal resolution, these signals provide real-time information on how the human body handles stress. Perez-Valero et al.[4] extracted the power density spectra of EEG signals to categorize stress and explored the effect of smoothing the power density spectra on the classification performance. Jebelli et al.[5] developed an online multitask learning algorithm to analyze stress in near real-time by using EEG features in time and frequency domains as well as frontal EEG asymmetry. Wen et al.[6] extracted the power densities of the Theta, Alpha, and Beta bands of EEG to classify the stress and suggested that the absolute power of the Beta band of the right frontal lobe is an effective feature for studying the stress. In traditional machine learning models for ECG-based stress classification, the features related to heart rate variability (HRV) are widely used in various studies[7-9].\nIn addition to traditional machine learning models, deep learning based EEG and ECG stress classification is also becoming popular. Mane et al.[10] designed a hybrid network of Convolution Neural Network(CNN) and Recurrent Neural Network(RNN) based on EEG signals to binary classify heart stress. Alruily et al.[11] used CNN and Long Short-Term Memory(LSTM) to extract features and combined with the Artificial Bee Colony (ABC) method and African buffalo optimization to classify acute stress and chronic stress based on EEG. Wang et al.[12] designed a convolutional network based on EEG to identify workers' stress in real-time. Giannakakis et al[13] investigated a 1D convolutional network based on ECG to classify psychological stress. Ishaque et al.[14] converted 1D ECG data to 2D images based on migration learning using the CNN network to apply the model to low performance devices to detect psychological stress, which verified the feasibility of the deep learning approach to detect stress in low performance devices.\nAlthough the above studies reported acceptable classification performance, they may still not be applicable in practical application scenarios, including factors such as insufficient classification accuracy and insufficient objective data. Multimodal-based studies can effectively address the above issues, such as the combination of ECG and EEG. Attar et al.[15] investigated the relevant features of EEG and ECG by comparing them through statistical analysis and reported features that were statistically significantly different for different levels of stress. Gonzalez-Carabarin et al. [16] investigated the effective features of EEG and ECG signals for stress classification respectively, reporting the most relevant EEG channels for different stress levels and the availability of HRV for stress analysis. Hemakom et al.[17] confirmed that multimodal approaches are more reliable compared to unimodal approaches by extracting features with high EEG and ECG correlations to input into a traditional machine learning model for stress classification. However, these methods rely on manual feature extraction, which may lead to the loss of key features, and it is time-consuming and laborious to try to find effective features, while deep learning multimodal fusion methods based on the stress classification task have rarely been studied.\nIn this study, we gathered data on four levels of psychological stress through the implementation of a mental arithmetic task experiment. We introduce a deep learning model that utilizes EEG and ECG bimodality for the purpose of studying stress categorization. Additionally, we will assess the potential of our multimodal deep learning model by comparing it to conventional machine learning methods.\nChapter 2 will detail the data collection procedure and the analysis techniques employed, while Chapter 3 will present the model's efficacy and compare it to other approaches. Finally, the study will conclude with a summary, discussion, and outlook.\n"},{"header":"2. Materials and Methods","content":"\u003ch2\u003e2.1 Data Acquisition\u003c/h2\u003e\n\u003cp\u003eNumerous studies have demonstrated that mental arithmetic tasks[18-20] can induce psychological stress, which will be utilized to induce stress in the current experiment. 26 volunteers, aged between 20-30, who were well-educated, physically and mentally healthy, and not abusing any substances, were recruited for the study. Informed consent forms were completed by all participants before the experiment, and the experimental protocol was in accordance with the Declaration of Helsinki.\u003c/p\u003e\n\u003cp\u003eEEG data were collected using a 14-channel wireless EMOTIV EpocX headset with a sampling frequency of 128 HZ. Figure 1 displays the electrode schematic.\u003c/p\u003e\n\u003cp\u003eECG data were gathered with a single-channel CHERO ECG patch at a sampling frequency of 250 HZ. The software for the experiment was designed using Python3.9 and Pyside6 framework on the Windows 10 platform. The computer controlled the marking signals for the data collection to ensure high temporal precision.\u003c/p\u003e\n\u003cp\u003eThe study is composed of five distinct phases: a resting phase, a practice phase, and three formal phases. Firstly, we will gather two minutes of resting state data from each subject, during which they are not required to perform any task but simply remain relaxed. A flow chart depicting the experiment for all phases except the resting phase is included in Figure 2.\u003c/p\u003e\n\u003cp\u003eEach stage comprises \u003cem\u003en\u0026nbsp;\u003c/em\u003esubsections, wherein participants are required to complete \u003cem\u003em\u003c/em\u003e questions within 10 seconds. The questions consist of three two-digit numbers, which are added or subtracted. The computer generates a random result that is close to the correct answer, and subjects are expected to choose whether the result is greater than, less than, or equal to the correct answer using the keyboard. There is a rest period of 5 seconds after each subsection and a 20-second rest period in the middle of each stage. The task difficulty increases with each stage. Ideally, no data indicating stress levels will be collected from the subjects during the resting stage. In stage 1, only light stress data will be collected, while in stage 2, concentrated stress data will be gathered. Heavy stress data will be collected during stage 3. The data obtained from the practicing stage and the resting time will be discarded. During the experiment, participants were instructed to minimize signal artifacts by refraining from making large movements and only applying slight pressure with their fingers on the keyboard. Failure to adhere to these guidelines resulted in the experiment being restarted or terminated, depending on the circumstances.\u003c/p\u003e\n\u003cp\u003eAfter assessing the data quality of 26 participants, we removed two subjects due to excessive data noise and preserved the data from the remaining 24 participants. For each subject, we selected the central 100 seconds of the resting state data over a 2-minute period and partitioned it into ten equal segments. Additionally, we obtained the data from each subsection of the three phases, yielding a total of 4*10=40 data samples. Each sample included 10 seconds of EEG and ECG data, with 14*1280 sampling points for EEG and 1*2500 sampling points for ECG. We collected a total of 24*40 data samples for 24 subjects, resulting in 960 overall samples.\u003c/p\u003e\n\u003ch2\u003e2.2 Data Pre-processing\u003c/h2\u003e\n\u003cp\u003eSince both EpocX and CHERO devices have some denoising of the signal and the denoising method used is not available, this paper will not take any pre-processing measures and will directly use the raw data for analysis to maximize the retention of information.\u003c/p\u003e\n\u003ch2\u003e2.3 Traditional Machine Learning Method\u003c/h2\u003e\n\u003ch3\u003e2.3.1 Feature Extraction\u003c/h3\u003e\n\u003cp\u003eEEG power band and power asymmetry in brain regions appear to have a strong correlation with psychological stress[4-6, 12]. The current study extracted the frequency band power of Delta (0.5-3.5 Hz), Theta (4-7.5 Hz), Alpha (8-13 Hz), and Beta (14-30 Hz) from 14 EEG channels. Furthermore, the frequency band power of seven symmetrical groups of channels (AF3-AF4, F3-F4, Fc5-FC6, F7-F8, T7-T8, P7-P8, O1-O2, shown in Figure 1) was analyzed for Alpha frequency band power asymmetry, as illustrated in Table 1.\u003c/p\u003e\n\u003cp\u003eTable 1. Description and numbers of extracted EEG features. Totally 63 EEG features in each sample.\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"796\" height=\"302\"\u003e\u003c/p\u003e\n\u003cp\u003eHRV, which responds to autonomic nervous system activity, has been the primary method for analyzing ECG signals[7-9]. The study presents the time domain, frequency domain, and statistical features extracted in Table 2.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTable\u0026nbsp;\u003c/em\u003e\u003cem\u003e2\u003c/em\u003e\u003cem\u003e. Description of extracted ECG features. Totally 12 features in each sample.\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"565\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003eFeature\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eVLF\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003epower\u0026nbsp;of\u0026nbsp;very\u0026nbsp;low\u0026nbsp;frequency(\u0026lt;0.04Hz)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eLF\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003epower\u0026nbsp;of\u0026nbsp;low\u0026nbsp;frequency(0.04-0.15Hz)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eHF\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003epower\u0026nbsp;of\u0026nbsp;high\u0026nbsp;frequency(0.15-0.4Hz)\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eLF/HF\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003eRatio of LF and HF\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eRMSSD\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003eroot\u0026nbsp;mean\u0026nbsp;of\u0026nbsp;squared\u0026nbsp;differences\u0026nbsp;of\u0026nbsp;successive\u0026nbsp;NN\u0026nbsp;Intervals\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eSDNN\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003estandard\u0026nbsp;deviation\u0026nbsp;of\u0026nbsp;an\u0026nbsp;NN\u0026nbsp;intervals\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eNN_mean\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003eMean\u0026nbsp;NN\u0026nbsp;interval\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eNN_min\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003eMinimum\u0026nbsp;NN\u0026nbsp;interval\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eNN_max\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003eMaximum\u0026nbsp;NN\u0026nbsp;interval\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eSDSD\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003estandard\u0026nbsp;deviation\u0026nbsp;of\u0026nbsp;differences\u0026nbsp;of\u0026nbsp;successive\u0026nbsp;NN\u0026nbsp;intervals\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003ePNN20\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003eNN20\u0026nbsp;count\u0026nbsp;divided\u0026nbsp;by\u0026nbsp;the\u0026nbsp;total\u0026nbsp;number\u0026nbsp;of\u0026nbsp;all\u0026nbsp;NN\u0026nbsp;intervals\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.70921985815603%\"\u003e\n \u003cp\u003e\u003cem\u003eTriangular\u0026nbsp;Index\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"74.29078014184397%\"\u003e\n \u003cp\u003e\u003cem\u003etriangular index based on the NN intervals histogram\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch3\u003e2.3.2 Classification\u003c/h3\u003e\n\u003cp\u003eSupport Vector Machines (SVMs), Decision Trees (DTs), Logistic Regression (LR), Naive Bayes (NB), K-Nearest Neighbors (KNN), and Stochastic Gradient Descent (SGD) algorithms are utilized for comparing classifications. The classification will be separated into two sessions. Firstly, the EEG and ECG features will be separately introduced to the classification algorithms to demonstrate the unimodal classification results. Subsequently, the multimodal fusion classification results will be presented utilizing feature fusion and decision fusion strategies, respectively.\u003c/p\u003e\n\u003cp\u003eIn the feature fusion approach, EEG and ECG features are overlayed and fed into the classifier. In the decision fusion method, both types of features are fed into the same classifier, and the outputs are added to generate the final classification result. Please refer to Figure 3 for the visual representation of these two techniques.\u003c/p\u003e\n\u003ch2\u003e2.4 Deep Learning Method\u003c/h2\u003e\n\u003cp\u003eCNN has demonstrated a high level of effectiveness in EEG and ECG signal-related tasks [10-14]. In this study, we utilized models based on CNN to categorize EEG and ECG signals separately, and subsequently incorporated the classification outcomes of both into a decision model for conclusive decision making. Next, we will present EEG, ECG, and categorical decision models in order.\u003c/p\u003e\n\u003ch3\u003e2.4.1 Model of EEG Classification\u003c/h3\u003e\n\u003cp\u003eFor EEG signals, we used the convolutional model shown in Figure 4:\u003c/p\u003e\n\u003cp\u003eThe initial 14-lead EEG signal serves as the model\u0026apos;s input and the corresponding data is allotted to the 14 channels of the convolutional layer. Temporal features are then obtained from the Convolution1 Layer, consisting of 28 convolutional kernels with each one measuring (1,3) in size. Next, the first feature dimension (channel) is switched with the last dimension (sampling point). And the feature shape is transformed from (28, 1, 1278) to (1278, 1, 28). Then, the transposed feature is inputted into the Convolution2 Layer, which has 28 convolution kernels with dimensions of (1, 1) to further extract the features of each channel in the Convolution1 Layer. Subsequently, the transposition is reversed to the same dimensions of the feature as the output of Convolution1. Input the feature into the Convolution3 Layer, which contains 56 convolution kernels, each with dimensions of (1,3), to extract deep temporal features. Then, use the channel maximum pooling layer to downscale the feature dimensions from (56,1,1276) to (1,1,1276). Finally, flatten the feature and input it into the Full Connection Layer for classification into four classes. Notably, the convolution process does not involve padding.\u003c/p\u003e\n\u003ch3\u003e2.4.2 Model of ECG Classification\u003c/h3\u003e\n\u003cp\u003eInspired by ST-CNN-GAP-5-Net[21], we designed a convolutional network to analyze ECG signals as shown in Figure 5.\u003c/p\u003e\n\u003cp\u003eThe model comprises six convolution modules, and Figure 5 displays the structure of each module in chronological order: Convolution Layer, Batch Normalization Layer, ReLu Layer, and Pooling Layer. The first five modules are Temporal Convolution Modules, while the last one is a Spatial Convolution Module. The kernel size for each filter is (1,5), (1,5), (1,5), (1,5), (1,3), and (1,3), respectively. Additionally, pooling is applied with sizes of (1,2), (1,4), (1,2), (1,4), (1,2), and (1,4). The original ECG signal is processed through various filters, including 4, 8, 16, 32, and 64. The Spatial Convolution Module is then used to extract spatial information from the features. Next, the features undergo downsizing through channel convolution with 64 filters and a kernel size of (12, 1). Lastly, global average pooling is employed. Finally, the extracted features are inputted into the Dense Module, which includes a Full Connection Layer, Batch Normalization Layer, ReLu Layer, and Dropout Layer (dropout rate of 0.1), to generate predicted probabilities for the four classes. Padding is applied throughout the convolution process.\u003c/p\u003e\n\u003ch3\u003e2.4.3 Model of Decision Fusion\u003c/h3\u003e\n\u003cp\u003eThe predicted probabilities of EEG and ECG for the four categories were acquired from the models mentioned in Sections 2.4.2 and 2.4.3, respectively. Then, all the probable values of the samples in the training set were inserted into the decision fusion model presented in Figure 6 for \u0026nbsp;training.\u003c/p\u003e\n\u003cp\u003eFirst, the predicted probabilities of EEG and ECG for the four pressure levels are added separately to calculate a new probability. This probability is then combined with the predicted probabilities of EEG and ECG to create a decision matrix with a size of (4,3). The matrix is inputted into the Decision Convolution Model illustrated in Fig. 6. The Convolution1 Layer, with filters of 2 and kernel size of (2,3), will learn the probabilities of EEG, ECG, and their sum to automatically assign weights. Then, the Convolution2 Layer with 4 filters and kernel size (2,1), will extract the lateral decision features again, flatten all features, and input them into the Full Connection Layer to obtain the final classification result. Padding is not used throughout the entire convolution process.\u003c/p\u003e\n\u003ch2\u003e2.5 Performance Metrics\u003c/h2\u003e\n\u003cp\u003eAccuracy and Kappa score are commonly used as a measure of model performance and are also considered to be the most important in classification tasks, and will serve as the basis for comparing model performance in this paper.\u003c/p\u003e\n\u003cp\u003eAccuracy is the ratio of correct predictions to total predicted values. To calculate accuracy, it is necessary to first calculate true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) using the following formula:\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"296\" height=\"58\"\u003e\u003c/p\u003e\n\u003cp\u003eThe Kappa coefficient calculates whether the model\u0026apos;s prediction accuracy is balanced for each class and is determined by:\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"271\" height=\"70\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere \u003cem\u003ePa\u003c/em\u003e is the actual percentage of agreement, and\u003cem\u003e\u0026nbsp;Pe\u003c/em\u003e is the expected percentage chance of agreement.\u003c/p\u003e\n\u003ch2\u003e2.6 1D-Grad-CAM\u003c/h2\u003e\n\u003cp\u003eGrad-CAM[22] obtains the weight of any convolutional layer on the target category through backpropagation. The original method displays features obtained by the convolutional layer on the input image. This paper combines the weights of the convolutional layer with the input signal and displays information obtained by the convolutional layer in the 1D signal for improved interpretability. For the principle of Grad-CAM derivation, please refer to the original article.\u003c/p\u003e"},{"header":"3. Experiments and Results","content":"\u003cp\u003eAll models were trained and tested on the stressed dataset presented in Chap.\u0026nbsp;2. The models were compared using a 5-fold cross-validation and Leave Three Subject Out (LTSO) validation method. PyTorch 1.8, Scikit-learn 1.2, and Tensorflow 2.6 were utilized to construct the modeling framework, with NVIDIA GTX 1070 GPU being utilized to expedite the model training. The optimizer used was Adam.\u003c/p\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Leave Three Subjects Out (LTSO)\u003c/h2\u003e\n \u003cp\u003eThe approach employed is derived from the Leave One Subject Out (LOSO) method. As this experiment only has 40 samples from one subject out of a total of 24 subjects, the amount of data is insufficient for the test set. Therefore, this thesis retains the data of three subjects for the test set, while the data of the remaining 21 subjects is used for the training set. To consider both the experiment\u0026apos;s quality and time constraints, we randomly divided the 24 subjects into randomized groups, with three in each group and a total of eight groups. We then used the data from one group as a test set and the other seven as a train set, similar to an 8-fold cross-validation approach.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Results of Traditional Machine Learning Method\u003c/h2\u003e\n \u003cp\u003eIn this section, we present the results of classifying EEG and ECG individually and fusing them using conventional machine learning techniques.\u003c/p\u003e\n \u003cp\u003eThe 63 previously mentioned EEG features and 12 ECG features undergo normalization using sklearn\u0026apos;s StandardScaler and are then utilized as inputs for classification by SVM, KNN, NB, LR, SGD, and DT classifiers. The outcomes of individual classification results will be contrasted with those produced by feature and decision fusion, after which the average accuracy(%) for both the five-fold cross-validation and the LTSO experiments will be displayed in Tables \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of the accuracy of five-fold cross-validation experiments based on traditional machine learning methods, with the best results indicated in bold.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassifier\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEEG only\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eECG only\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFeature fusion\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDecision fusion\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e39.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e36.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e43.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e44.89\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e45.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e44.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e46.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e55.52\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e26.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e27.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e37.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e34.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e41.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e37.50\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSGD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e35.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e34.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e34.06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e66.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e53.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e68.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e74.17\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of the accuracy of LTSO experiments based on traditional machine learning methods, with the best results indicated in bold.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassifier\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEEG only\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eECG only\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFeature fusion\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDecision fusion\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e39.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e37.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e40.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e42.08\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e51.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e61.25\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e26.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e29.58\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e38.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e34.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e42.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e38.85\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSGD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e33.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e26.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e36.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e31.45\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e71.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e58.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e72.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e82.18\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eThe results reveal that the fusion model yields higher classification accuracy than either EEG or ECG alone in both 5-fold cross-validation and LTSO experiments. This underscores the strengths of multimodal fusion. Furthermore, decision fusion outperforms feature fusion, indicating the former\u0026apos;s greater potential. Among the six classifiers, the DT classifier exhibits superior performance, leading with classification accuracies of 74.17% and 82.18% in 5-fold cross-validation and LTSO experiments, respectively.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\n \u003ch2\u003e3.3 Results of The Proposed Deep Learning Method\u003c/h2\u003e\n \u003cp\u003eThis section first compares the outcomes of the proposed models on an individual basis with those of the leading models in the same field. Afterward, it analyzes the results of the multimodal fusion models as compared to the unimodal ones.\u003c/p\u003e\n \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\n \u003ch2\u003e3.3.1 Comparison of EEG Models\u003c/h2\u003e\n \u003cp\u003eThe proposed EEG model was compared with DeepConvNet[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e], EEGNet[\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e], EEGNeX[\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e], ATCNet[\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e], and the best results (DT) obtained from the previously traditional machine learning model. Table \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e and Table \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e demonstrate the average accuracy and kappa score of the 5-fold cross-validation and LTSO experiments.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab5\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of classification results from 5-fold cross-validation experiments of the EEG models, with the best results bolded.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAccuracy(%)\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eKappa score\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e66.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.5547\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDeepConvNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e81.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7475\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEEGNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e80.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7447\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEEGNeX\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e86.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.8176\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eATCNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e87.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.8322\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed\u0026nbsp;EEG\u0026nbsp;model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e88.58\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8592\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab6\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of classification results from LTSO experiments of the EEG models, with the best results bolded.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAccuracy(%)\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eKappa score\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e71.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.6236\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDeepConvNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e77.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.6972\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEEGNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e80.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7416\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEEGNeX\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e84.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7886\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eATCNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e85.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7992\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed\u0026nbsp;EEG\u0026nbsp;model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e85.86\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8111\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eThe findings indicate that deep learning approaches typically outperform traditional machine learning approaches, and the proposed EEG model achieves the best classification performance when used for stress classification.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e\n \u003ch2\u003e3.3.2 Results of The Proposed ECG Model\u003c/h2\u003e\n \u003cp\u003eMost studies on ECG stress classification focused on analyzing HRV, and only a few have utilized deep learning. Furthermore, few models with open-source code are available. This section presents the ablation study of the proposed model and compares it with the DT method, which was introduced previously. The ablation experiments determine the effects of decreasing or increasing the temporal convolutional module on the classification. The results of the 5-fold cross-validation and LTSO experiments are presented in Tables \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e, respectively.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab7\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of classification results from 5-fold cross-validation experiments of ECG model, with the best results bolded, where 4-T-conv-layer and 6-T-conv-layer mean 4 and 6 Temporal Convolution Layer utilized in proposed ECG model.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAccuracy(%)\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eKappa score\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e53.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.3739\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4-T-conv-layer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e63.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.5158\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6-T-conv-layer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e82.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7724\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed\u0026nbsp;ECG\u0026nbsp;model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e85.21\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8023\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab8\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of classification results from LTSO experiments of ECG model, with the best results bolded, where 4-T-conv-layer and 6-T-conv-layer mean 4 and 6 Temporal Convolution Layer utilized in proposed ECG model.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAccuracy(%)\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eKappa score\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e58.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.4472\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4-T-conv-layer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e66.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.5367\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6-T-conv-layer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e83.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7836\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed\u0026nbsp;ECG\u0026nbsp;model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e87.29\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8305\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eThe findings indicated that the ECG deep learning model proposed yields much higher performance than traditional machine learning methods. Additionally, the proposed 5-layer temporal convolution module outperformed both the 4-layer temporal convolution module and the 6-layer convolution module in terms of classification performance.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e\n \u003ch2\u003e3.3.3 Comparison of Multimodal and Unimodal Methods\u003c/h2\u003e\n \u003cp\u003eComparison of classification results of multimodal and unimodal decision fusion models are shown in Tables \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e as follows.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab9\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 9\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of classification results from multimodal and unimodal 5-fold cross-validation experiments, with the best results shown in bold.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAccuracy(%)\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eKappa score\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEEG Only\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e88.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.8592\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eECG Only\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e85.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.8023\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed Fusion Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e91.14\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8817\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab10\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 10\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of classification results from multimodal and unimodal LTSO experiments, with the best results shown in bold.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAccuracy(%)\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eKappa score\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEEG Only\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e85.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.8111\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eECG Only\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e87.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.8305\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed Fusion Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e91.97\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8931\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eThe results demonstrated that the multimodal fusion model outperforms the unimodal model in terms of classification performance. Additionally, decision fusion enhanced classification performance by automatically assigning weights through the convolutional network, particularly in situations where there was an imbalance in the classification effect of EEG and ECG models.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\n \u003ch2\u003e3.3.4 Feature Visualization\u003c/h2\u003e\n \u003cp\u003eIn this section, we utilized the 1D-Grad-CAM technique to visualize the proposed convolutional network model and display the features acquired by the convolutional layers which could enhance the interpretability of the model.\u003c/p\u003e\n \u003cp\u003eThe second and third convolutional layers of the EEG model will be fed into the 1D Grad-CAM to demonstrate the selected features, and we randomly selected one correctly classified sample from each class of all samples visualized in Fig. \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e:\u003c/p\u003e\n \u003cp\u003eThe results indicated that the second convolutional layer of the EEG model prioritizes continuous EEG signal features, while the third convolutional layer was predominantly focused on lower frequency information. This finding is consistent with the majority of studies that manually extract features, including lower frequency band power.\u003c/p\u003e\n \u003cp\u003eThe temporal convolutions of the ECG model from the second through the fifth layers, along with the final spatial convolution, were utilized for feature visualization in the 1D Grad-CAM, as illustrated in Fig. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eThe results indicated that the first three layers of the temporal convolution concentrate on features primarily within the QRS wave range, whereas the fifth layer concentrates on global features, and the spatial convolution focuses on lower frequency data, consistent with the commonly used HRV analysis for ECG. Confirming the model\u0026apos;s feature-extraction effectiveness.\u003c/p\u003e\n \u003cp\u003eThe last layer of convolutional features of the decision model was also visualized to show the different weights given to the EEG ECG decisions, four correctly identified samples of different categories of stress were randomly selected and the results are shown in Fig. \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e:\u003c/p\u003e\n \u003cp\u003eAs can be seen from the figure for no or low stress levels EEG is selected with more weight, while for moderate or severe stress the selection of ECG is more important.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"4. Discussion and Conclusion","content":"\u003cp\u003eIn this study, the performance difference between the traditional machine learning method of manual feature extraction and the convolution-based deep learning method is investigated by collecting EEG and ECG data from 24 subjects with four types of stress levels, which confirms the superior performance of the proposed unimodal model among the existing methods, and the deep learning-based decision fusion model is investigated on the basis of the unimodal model, which further improves the classification performance of the model and confirms the effectiveness of the multimodal model in the pressure detection task. In addition, the features extracted from the convolutional layer are visualized using the Grad-CAM method, which enhances the interpretability of the model.\u003c/p\u003e \u003cp\u003eFrom the results, the experimental LTSO accuracy of ECG is higher than 5-fold cross-validation, and the experimental LTSO accuracy of EEG is lower than 5-fold cross-validation, which means that the intersubject variability of ECG is lower than that of EEG, which may be related to the low signal-to-noise ratio of EEG.\u003c/p\u003e \u003cp\u003eAlthough acceptable classification performance was achieved in this study, the proposed decision fusion model proposed in this study failed to address the degradation of classification performance caused when the difference in unimodal classification performance is too large, for example, at the beginning of the study, a different ECG model was applied for stress classification, but only about 78% accuracy was achieved, which was about 10% different from the accuracy of the EEG model, resulting in the classification accuracy of the decision fusion model failing to exceed that using only EEG. In the future, we will try to overcome this limitation with improved attention mechanisms.\u003c/p\u003e \u003cp\u003eOverall, to address the lack of performance in stress classification tasks, deep learning based decision fusion models have great potential.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate:\u003c/strong\u003e Not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e: Not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials:\u0026nbsp;\u003c/strong\u003eThe data that support the findings of this study are available on request from the author, upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u0026nbsp;\u003c/strong\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003eThis research was funded by Department of Science and Technology of Shandong Province, grant number ZR2020QF018.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u003c/strong\u003e Conceptualization, B.Z. and L.W.; methodology, B.Z.; software, B.Z.; validation, L.W. and C.J.; writing\u0026mdash;original draft preparation, B.Z.; funding acquisition, L.W. and C.J.; All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgement:\u003c/strong\u003e Not applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eLevine, G. N. Psychological Stress and Heart Disease: Fact or Folklore? The American Journal of Medicine 2022, 135 (6), 688\u0026ndash;696. https://doi.org/10.1016/j.amjmed.2022.01.053.\u003c/li\u003e\n\u003cli\u003eHammen, C. Stress and Depression. Annual Review of Clinical Psychology 2005, 1 (1), 293\u0026ndash;319. https://doi.org/10.1146/annurev.clinpsy.1.102803.143938.\u003c/li\u003e\n\u003cli\u003eSinha, R. Chronic Stress, Drug Use, and Vulnerability to Addiction. Ann N Y Acad Sci 2008, 1141, 105\u0026ndash;130. https://doi.org/10.1196/annals.1441.030.\u003c/li\u003e\n\u003cli\u003ePerez-Valero, E.; Lopez-Gordo, M. A.; Vaquero-Blasco, M. A. EEG-Based Multi-Level Stress Classification with and without Smoothing Filter. Biomedical Signal Processing and Control 2021, 69, 102881. https://doi.org/10.1016/j.bspc.2021.102881.\u003c/li\u003e\n\u003cli\u003eJebelli, H.; Mahdi Khalili, M.; Lee, S. A Continuously Updated, Computationally Efficient Stress Recognition Framework Using Electroencephalogram (EEG) by Applying Online Multitask Learning Algorithms (OMTL). IEEE Journal of Biomedical and Health Informatics 2019, 23 (5), 1928\u0026ndash;1939. https://doi.org/10.1109/JBHI.2018.2870963.\u003c/li\u003e\n\u003cli\u003eWen, T. Y.; Mohd Aris, S. A. Hybrid Approach of EEG Stress Level Classification Using K-Means Clustering and Support Vector Machine. IEEE Access 2022, 10, 18370\u0026ndash;18379. https://doi.org/10.1109/ACCESS.2022.3148380.\u003c/li\u003e\n\u003cli\u003eVanitha, L.; Suresh, G. R. Hybrid SVM Classification Technique to Detect Mental Stress in Human Beings Using ECG Signals. In 2013 International Conference on Advanced Computing and Communication Systems; 2013; pp 1\u0026ndash;6. https://doi.org/10.1109/ICACCS.2013.6938735.\u003c/li\u003e\n\u003cli\u003ePourmohammadi, S.; Maleki, A. Stress Detection Using ECG and EMG Signals: A Comprehensive Study. Computer Methods and Programs in Biomedicine 2020, 193, 105482. https://doi.org/10.1016/j.cmpb.2020.105482.\u003c/li\u003e\n\u003cli\u003eKuttala, R.; Subramanian, R.; Oruganti, V. R. M. Hierarchical Autoencoder Frequency Features for Stress Detection. IEEE Access 2023, 11, 103232\u0026ndash;103241. https://doi.org/10.1109/ACCESS.2023.3316365.\u003c/li\u003e\n\u003cli\u003eMane, S. A. M.; Shinde, A. StressNet: Hybrid Model of LSTM and CNN for Stress Detection from Electroencephalogram Signal (EEG). Results in Control and Optimization 2023, 11, 100231. https://doi.org/10.1016/j.rico.2023.100231.\u003c/li\u003e\n\u003cli\u003eAlruily, M. Sentiment Analysis for Predicting Stress among Workers and Classification Utilizing CNN: Unveiling the Mechanism. Alexandria Engineering Journal 2023, 81, 360\u0026ndash;370. https://doi.org/10.1016/j.aej.2023.09.040.\u003c/li\u003e\n\u003cli\u003eWang, Y.; Huang, Y.; Gu, B.; Cao, S.; Fang, D. Identifying Mental Fatigue of Construction Workers Using EEG and Deep Learning. Automation in Construction 2023, 151, 104887. https://doi.org/10.1016/j.autcon.2023.104887.\u003c/li\u003e\n\u003cli\u003eGiannakakis, G.; Trivizakis, E.; Tsiknakis, M.; Marias, K. A Novel Multi-Kernel 1D Convolutional Neural Network for Stress Recognition from ECG. In 2019 8th International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW); 2019; pp 1\u0026ndash;4. https://doi.org/10.1109/ACIIW.2019.8925020.\u003c/li\u003e\n\u003cli\u003eIshaque, S.; Khan, N.; Krishnan, S. Detecting Stress through 2D ECG Images Using Pretrained Models, Transfer Learning and Model Compression Techniques. Machine Learning with Applications 2022, 10, 100395. https://doi.org/10.1016/j.mlwa.2022.100395.\u003c/li\u003e\n\u003cli\u003eAttar, E. T.; Balasubramanian, V.; Subasi, E.; Kaya, M. Stress Analysis Based on Simultaneous Heart Rate Variability and EEG Monitoring. IEEE Journal of Translational Engineering in Health and Medicine 2021, 9, 1\u0026ndash;7. https://doi.org/10.1109/JTEHM.2021.3106803.\u003c/li\u003e\n\u003cli\u003eGonzalez-Carabarin, L.; Castellanos-Alvarado, E. A.; Castro-Garcia, P.; Garcia-Ramirez, M. A. Machine Learning for Personalised Stress Detection: Inter-Individual Variability of EEG-ECG Markers for Acute-Stress Response. Computer Methods and Programs in Biomedicine 2021, 209, 106314. https://doi.org/10.1016/j.cmpb.2021.106314.\u003c/li\u003e\n\u003cli\u003eHemakom, A.; Atiwiwat, D.; Israsena, P. ECG and EEG Based Detection and Multilevel Classification of Stress Using Machine Learning for Specified Genders: A Preliminary Study. PLOS ONE 2023, 18 (9), e0291070. https://doi.org/10.1371/journal.pone.0291070.\u003c/li\u003e\n\u003cli\u003eHe, J.; Li, K.; Liao, X.; Zhang, P.; Jiang, N. Real-Time Detection of Acute Cognitive Stress Using a Convolutional Neural Network From Electrocardiographic Signal. IEEE Access 2019, 7, 42710\u0026ndash;42717. https://doi.org/10.1109/ACCESS.2019.2907076.\u003c/li\u003e\n\u003cli\u003eMcDuff, D.; Gontarek, S.; Picard, R. Remote Measurement of Cognitive Stress via Heart Rate Variability. In 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society; 2014; pp 2957\u0026ndash;2960. https://doi.org/10.1109/EMBC.2014.6944243.\u003c/li\u003e\n\u003cli\u003eGiannakakis, G.; Grigoriadis, D.; Giannakaki, K.; Simantiraki, O.; Roniotis, A.; Tsiknakis, M. Review on Psychological Stress Detection Using Biosignals. IEEE Transactions on Affective Computing 2022, 13 (1), 440\u0026ndash;460. https://doi.org/10.1109/TAFFC.2019.2927337.\u003c/li\u003e\n\u003cli\u003eAnand, A.; Kadian, T.; Shetty, M. K.; Gupta, A. Explainable AI Decision Model for ECG Data of Cardiac Disorders. Biomedical Signal Processing and Control 2022, 75, 103584. https://doi.org/10.1016/j.bspc.2022.103584.\u003c/li\u003e\n\u003cli\u003eSelvaraju, R. R.; Cogswell, M.; Das, A.; Vedantam, R.; Parikh, D.; Batra, D. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization. In 2017 IEEE International Conference on Computer Vision (ICCV); 2017; pp 618\u0026ndash;626. https://doi.org/10.1109/ICCV.2017.74.\u003c/li\u003e\n\u003cli\u003eSchirrmeister, R. T.; Springenberg, J. T.; Fiederer, L. D. J.; Glasstetter, M.; Eggensperger, K.; Tangermann, M.; Hutter, F.; Burgard, W.; Ball, T. Deep Learning with Convolutional Neural Networks for EEG Decoding and Visualization. Hum Brain Mapp 2017, 38 (11), 5391\u0026ndash;5420. https://doi.org/10.1002/hbm.23730.\u003c/li\u003e\n\u003cli\u003eLawhern, V. J.; Solon, A. J.; Waytowich, N. R.; Gordon, S. M.; Hung, C. P.; Lance, B. J. EEGNet: A Compact Convolutional Network for EEG-Based Brain-Computer Interfaces. J. Neural Eng. 2018, 15 (5), 056013. https://doi.org/10.1088/1741-2552/aace8c.\u003c/li\u003e\n\u003cli\u003eChen, X.; Teng, X.; Chen, H.; Pan, Y.; Geyer, P. Toward Reliable Signals Decoding for Electroencephalogram: A Benchmark Study to EEGNeX. Biomedical Signal Processing and Control 2024, 87, 105475. https://doi.org/10.1016/j.bspc.2023.105475.\u003c/li\u003e\n\u003cli\u003eAltaheri, H.; Muhammad, G.; Alsulaiman, M. Physics-Informed Attention Temporal Convolutional Network for EEG-Based Motor Imagery Classification. IEEE Transactions on Industrial Informatics 2023, 19 (2), 2249\u0026ndash;2258. https://doi.org/10.1109/TII.2022.3197419.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Multimodel Fusion, Convolution Neural Network(CNN), Stress Classification, Electroencephalogram(EEG), Electrocardiogram(ECG), Decision Fusion, 1D-Grad-CAM","lastPublishedDoi":"10.21203/rs.3.rs-4015916/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4015916/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePsychological stress cannot be ignored in today's society, and there is an urgent need for an objective and cost-effective method to detect it. However, traditional machine learning methods that require manual feature extraction require a lot of research time and cannot guarantee accuracy. In this paper, we establish a four-category stress multimodal dataset by collecting EEG and ECG signals from 24 subjects performing mental arithmetic tasks with different difficulty levels and propose a multimodal decision fusion model based on Convolution Neural Network to classify the data. The prediction probabilities of EEG and ECG signals for the four stress categories are first extracted by two models each and then fused into the decision model for the final classification, \u0026nbsp;5-fold cross-validation and Leave-Three-Subjects-Out experiments are performed, which achieve 91.14% and 91.97% accuracy, respectively. In addition, the features of the convolution layer were visualized using the 1D-Grad-CAM method to improve the interpretability of the model.\u003c/p\u003e","manuscriptTitle":"Psychological Stress Classification Using EEG and ECG: A CNN Based Multimodal Fusion Model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-12 10:57:12","doi":"10.21203/rs.3.rs-4015916/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"7d5e6e6b-0dfb-4675-a23e-2f869f2e1b79","owner":[],"postedDate":"March 12th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-05-01T07:53:30+00:00","versionOfRecord":[],"versionCreatedAt":"2024-03-12 10:57:12","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4015916","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4015916","identity":"rs-4015916","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}