Predicting Dyskinetic Events through Verified Multivariate Time Series Classification

Bergami, G; Packer, E; Scott, K; Del, Din, S

doi:10.1007/978-3-031-83472-1_4

Predicting Dyskinetic Events through Verified Multivariate Time Series Classification

Lookup NU author(s): Dr Giacomo Bergami ORCiD, Emma Packer, Dr Kirsty Scott ORCiD, Dr Silvia Del Din ORCiD

Downloads

Accepted version [.pdf]

Licence

This is the authors' accepted manuscript of a conference proceedings (inc. abstract) that has been published in its final definitive form by Springer Verlag, 2024.

For re-use rights please refer to the publisher's terms and conditions.

Abstract

While monitoring Parkinsonian patients with wearable sensors and tracking their drug assumption patterns, we want to differentiate the behaviours distinguishing periods of relative well-being from dyskinetic events. This requires solving a novel problem, where an entire multivariate time series (MTS) has its class label varying in time, thus leading to a generalised formulation of multivariate time series classification (MTSC). To achieve explainability, we premier the composition of data trend (DT) analysis with DECLAREd, a log temporal declarative language, to derive human-readable correlations across different MTS dimensions' trends. This is mediated by a novel temporal data representation, polyadic logs, supporting both MTS raw data and concurrent activity-labelled durative activities (constituents) for representing event-based classes and concurrent DTs across MTS dimensions. Our validation over a real patient dataset shows that our MTCS algorithm, EMeriTAte, outperforms state-of-the-art MTSC for a novel patient classification task.