Abstract
The concept of recurrence in nonlinear dynamics has been found useful for discovering patterns in complex time series of natural, physical, and biological processes. The method of fuzzy recurrence plots has recently been developed for studying patterns of recurrent behaviors in dynamical systems. Analysis of physiological time series has increasingly become important for medical research, and deep learning is reported in literature as the most advanced approach in artificial intelligence for classification of time series. For the first time, this paper presents the idea of computing texture properties of fuzzy recurrence of physiological time series to be used as input data for classification of physiological time series with deep recurrent neural networks. A public gait in Parkinson's disease database was used to test the performance of the proposed approach. The deep learning of texture can significantly increase improvements in classification accuracy over some existing deep-learning models.