A neuro-fuzzy model using physiological data during virtual reality therapy achieved an 83% mean recognition rate for stress level classification, outperforming other machine learning classifiers.
Does a neuro-fuzzy model analyzing physiological data improve stress level classification compared to standard machine learning algorithms in subjects undergoing virtual reality therapy?
A neuro-fuzzy model utilizing ECG and respiration data can accurately predict stress levels during virtual reality therapy, potentially enhancing physiological computing for stress treatment.
This paper reports the design and assessment of a neuro-fuzzy model to support clinicians during virtual reality therapy. The implemented model is able to automatically recognize the perceived stress levels of the patients by analyzing physiological and behavioral data during treatment. The model, consisting of a self-organizing map and a fuzzy-rule-based module, was trained unobtrusively recording electrocardiogram, breath rate and activity during stress inoculation provided by the exposure to virtual environments. Twenty nurses were exposed to sessions simulating typical stressful situations experienced at their workplace. Four levels of stress severity were evaluated for each subject by gold standard clinical scales administered by trained personnel. The model's performances were discussed and compared with the main machine learning algorithms. The neuro-fuzzy model shows better performances in terms of stress level classification with 83% of mean recognition rate. RESEARCH HIGHLIGHTS Stress levels were predicted on the basis of physiological computing using a neuro-fuzzy model during virtual reality therapy. Features were extracted from ECG and respiration obtaining high accuracy and optimization of computational costs. The neuro-fuzzy model shows better performance than the more frequently adopted classifiers. This approach may enhance the use of physiological computing for stress treatment in clinical practice.
Tartarisco et al. (Wed,) conducted a other in Stress (n=20). Neuro-fuzzy model for stress level classification vs. Main machine learning algorithms was evaluated on Stress level classification (mean recognition rate). A neuro-fuzzy model using physiological data during virtual reality therapy achieved an 83% mean recognition rate for stress level classification, outperforming other machine learning classifiers.