Machine Learning for personalised stress detection: Inter-individual variability of EEG-ECG markers for acute-stress response

Gonzalez-Carabarin, Lizeth; Castellanos-Alvarado, E.A.; Castro-Garcia, P.; García-Ramírez, Mario Alberto

doi:10.1016/j.cmpb.2021.106314

Cited by 24 publications

(19 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As seen in Figure 2 a, EDA was the signal most often used in the studies, followed by ECG and EEG. The studies were usually based only on EEG, and they generally did not combine EEG with other signals, except for Gonzalez-Carabarin et al [ 23 ], who combined EEG with ECG, and Xu et al [ 24 ], who combined EEG with ECG, EMG, and EDA. Figure 2 b shows the sample size by study according to the gender of the participants.…”

Section: Resultsmentioning

confidence: 99%

“…Two studies [ 23 , 27 ] used small sample sizes from a student population characterized as young, active, and healthy. One might therefore question the usefulness of studies such as this performed on a non-representative sample in real life.…”

Section: Discussion and Future Directionsmentioning

confidence: 99%

“…The two studies combining EEG with other signals first applied K-means [ 24 ] and then another cluster-wise classification [ 23 ]. On the one hand, Xu et al [ 24 ] used K-means to divide the subjects into different categories, and then regression analysis was conducted using the generalized regression neural network (GRNN) for individual clusters based on the training dataset belonging to the task load and recovery stages.…”

Section: Modelsmentioning

confidence: 99%

“…This led to a set of K GRNN models that minimized the cluster-wise error. On the other hand, Gonzalez-Carabarin et al [ 23 ] used unsupervised learning to cluster the EEG features into stress and non-stress periods (plus an extra cluster to categorize data points that were not part of previous categories) and then applied supervised learning. As seen in Table 1 , for the same experiment, classification into two categories was more accurate than classification into three categories; thus, it was still difficult to accurately classify levels of anxiety.…”

Section: Modelsmentioning

confidence: 99%

“…However, this result contradicts the results of Gazi et al [ 29 ], who found an accuracy of 78% with a combination of ECG and EDA and 85% with a combination of ECG, EDA, and RSP. A combination of skin temperature (ST), EDA, RSP, and HR led to the best performance using SVM, followed by a study [ 28 ] using only EEG and a study [ 23 ] that combined ECG and EEG. For neural networks, the best accuracy was obtained with the study [ 33 ] that combined EDA, PPG, and ST. Good performance corresponding to accuracy above 85% was also achieved with EEG or EEG combined with ECG, EDA, and EMG.…”

Section: Modelsmentioning

confidence: 99%

See 4 more Smart Citations

Machine Learning for Anxiety Detection Using Biosignals: A Review

2022

View full text Add to dashboard Cite

Anxiety disorder (AD) is a major mental health illness. However, due to the many symptoms and confounding factors associated with AD, it is difficult to diagnose, and patients remain untreated for a long time. Therefore, researchers have become increasingly interested in non-invasive biosignals, such as electroencephalography (EEG), electrocardiogram (ECG), electrodermal response (EDA), and respiration (RSP). Applying machine learning to these signals enables clinicians to recognize patterns of anxiety and differentiate a sick patient from a healthy one. Further, models with multiple and diverse biosignals have been developed to improve accuracy and convenience. This paper reviews and summarizes studies published from 2012 to 2022 that applied different machine learning algorithms with various biosignals. In doing so, it offers perspectives on the strengths and weaknesses of current developments to guide future advancements in anxiety detection. Specifically, this literature review reveals promising measurement accuracies ranging from 55% to 98% for studies with sample sizes of 10 to 102 participants. On average, studies using only EEG seemed to obtain the best performance, but the most accurate results were obtained with EDA, RSP, and heart rate. Random forest and support vector machines were found to be widely used machine learning methods, and they lead to good results as long as feature selection has been performed. Neural networks are also extensively used and provide good accuracy, with the benefit that no feature selection is needed. This review also comments on the effective combinations of modalities and the success of different models for detecting anxiety.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Discussion and Future Directionsmentioning

confidence: 99%