Multimodal physiological correlates and prediction of transcutaneous electrotactile perception thresholds

Electrotactile technology, as a non-invasive means of haptic feedback, has shown great potential in virtual reality, human–computer interaction and rehabilitation medicine. The perception threshold (PT) is a key metric for individualised calibration in electrotactile systems. However, it is strongly influenced by inter-individual physiological variability and the state of the skin–electrode interface. Existing studies often focus on a limited set of factors, making it difficult to characterise their combined contribution to threshold variability. To address this, we propose a multimodal data-driven framework for electrotactile PT assessment and prediction. Using a rapid threshold measurement paradigm, we constructed a dataset comprising 128 finger-level records from 32 healthy participants aged 21–27 years, integrating global physiological, local morphological and bioelectrical features. A nested cross-validation procedure used three inner folds for feature selection and parameter tuning and five outer folds for performance evaluation. Experimental results show that, for exploratory binary classification using the sample median of 156 μs as the decision boundary, the Random Forest achieved accuracies of 74.25% and 72.86% under non-cross-subject and cross-subject settings, respectively, exceeding the 50% baseline. The full multimodal model also achieved higher accuracy than single-modality models and reference models using sex alone or sex together with height. These findings identify associations between multimodal measurements and PT and support their predictive value within the studied cohort and stimulation configuration, providing a data foundation and modelling evidence for further investigation of individualised electrotactile calibration.