Emotional stress classification using spiking neural networks.

Abstract

Objective: This study examined the data modeling capability of spiking neural networks (SNN) in classifying stressed versus relaxed brain states using electroencephalogram (EEG) data. The input spatiotemporal dynamics were explored to obtain further knowledge regarding the two-brain states. Method: A publicly available EEG data set for emotion analysis using psychological signals (DEAP) collected from 32 participants (50% females) with an average age of 26.9 is used in this study. Firstly, data extraction is performed using a criterion that defines stress and relaxation states using self-reported valence and arousal scores. Two hundred eight such extracted samples were selected to train and evaluate a novel three-layer feedforward SNN. This SNN consisted of leaky-integrate and fire neurons and learned from incoming data using spike-time-dependent plasticity (STDP) and dynamically evolving SNN algorithms. The SNN performance was evaluated using both fivefold cross-validation and a 60:40 training testing split. To explore input spatiotemporal dynamics, a specialized SNN architecture for brain data processing named NeuCube was used. Results: The highest-performing model of the novel SNN algorithm produced 88% average accuracy (F1 score: 86.21%, Matthews correlation coefficient: 0.78). This SNN outperformed traditional machine learning (ML) techniques without the use of manual feature extraction. Moreover, the input dynamics revealed higher prefrontal activation during relaxation states compared to stress states. Conclusions: The results showed the capability of the SNN algorithm to recognize stressed and relaxed states of the brain, using temporal learning techniques. Furthermore, the findings obtained from NeuCube suggested a potential approach for brain data analysis, setting SNNs apart from black box approaches used for brain data processing.

Sequence Modeling of Passive Sensing Data for Treatment Response Prediction in Major Depressive Disorder

“…We discussed both recurrent neural networks and classic machine learning methods in the related work, but did not cover Convolutional Neural Networks (CNNs) or Spiking Neural Networks (SNNs). Indeed, these models have the potential to excel in sequence modeling tasks [51]- [53]. The primary distinction between a CNN and an RNN is their ability to process temporal information.…”

Mental stress recognition on the fly using neuroplasticity spiking neural networks

“…When considering the DEAP dataset, the O-NSNN could not outperform SNN and SVM techniques built for stress recognition (Table 3 ). The methods that outperformed the O-NSNN used feature engineering 61 or hyperparameter optimization 65 methods for the modelling tasks. Exploring the modelling mechanisms of O-NSNN, we found that the EDs of output neurons (i.e., numerical representations of input samples) to have better discriminative capability between the initial and final states of O-NSNN than in O-RSNN.…”

Neurobiological Link between Stress and Gaming: A Scoping Review

“…EEG signals containing information about mental states might be a helpful tool to measure changes in brain activity patterns caused by stress [41,42]. EEG asymmetry index, power, coherence, and other features were widely investigated in stress studies, with the alpha asymmetry index and greater frontal right alpha activity being a consistent and robust stress-related feature reflecting emotional arousal [42,43].…”