Surrogate Rehabilitative Time Series Data for Image-based Deep Learning

Big Data comprise the tools to analyse vast stores of data generated by the myriad of powerful, low-cost processors, sensors and networks around us. The spiralling demand for multi-sensored personal communication devices has played a major role in this, producing images and leaving digital trails of transactions and texts to be mined for patterns. Consequently, the cutting edge of data analysis tools has been targeted for images. However the copious amounts of data required to successfully train these tools are not available in several fields such as rehabilitation where there are constraints on data collection. And yet the need for timely clinical assessments grows.We consider how to address this situation by generating synthetic, surrogate data which preserves many properties of the original. Here we introduce a new application of surrogate time series in a novel classification scheme, compare methods of converting these into images and use a state of the art neural network framework for a successful improvement in classification results.This is a significant contribution to the art, demonstrating how scarce time series data can be successfully augmented to take advantage of cutting edge analytical tools.