Multivariate functional principal component analysis to predict exercise type in compression therapy

Nastaran Sharifian, University of Galway

Co-authors: Andrew Simpkin, University of Galway

Abstract: Venous leg ulcers (VLUs) are commonly treated with compression therapy; however, manual bandaging introduces variability in pressure application. Wearable sensors enable continuous monitoring of under-bandage pressure, generating large datasets that require advanced statistical methods. This study demonstrates the use of Multivariate Functional Principal Component Analysis (MFPCA) to effectively reduce the dimensionality of pressure wave data recorded simultaneously from multiple sensor positions, simplifying the classification of exercise types using multinomial logistic regression. The MFPCA-based model achieves high predictive accuracy by capturing correlations across sensor locations with fewer principal components compared to a combined univariate FPCA approach. Additionally, individual FPCA analyses highlight the sensor positions most informative for predicting exercise type. Future research will validate this approach on larger datasets to confirm reliability and clinical applicability.