Related Experiment Videos
Using echo state networks for classification: A case study in Parkinson's disease diagnosis.
Stuart E Lacy1, Stephen L Smith1, Michael A Lones2
1University of York, York, United Kingdom.
Artificial Intelligence in Medicine
|February 25, 2018
Summary
Echo state networks (ESNs) offer efficient time series analysis for disease prediction using movement data. These reservoir computing methods provide rapid classification without domain expertise, outperforming traditional approaches.
Area of Science:
- Computational neuroscience
- Machine learning
- Biomedical engineering
Background:
- Reservoir computing, including echo state networks (ESNs), presents advantages for time series analysis but lacks widespread adoption in data mining.
- Limited research exists on applying ESNs for classification tasks.
Purpose of the Study:
- To demonstrate the efficacy of ESNs for disease prediction using movement data.
- To explore various input-output mapping techniques for ESN classification.
- To highlight ESNs' potential in rapid, domain-knowledge-free predictive modeling.
Main Methods:
- Training echo state networks (ESNs) on movement data to predict disease labels.
- Investigating different approaches for input-output mapping in ESNs for classification.
- Utilizing wearable sensor data from Parkinson's disease patients performing movement tasks.
Main Results:
- ESNs achieved effective classification of disease labels from movement data.
- ESN performance was competitive with traditional methods, including those requiring extensive feature engineering and longer training times.
- ESNs demonstrated rapid training capabilities.
Conclusions:
- ESNs are a viable and efficient tool for disease prediction using time series data, particularly movement data.
- ESNs offer a powerful alternative for rapid model training without requiring domain expertise, suitable for high-dimensional wearable sensor data.
- The study validates ESNs for applications like Parkinson's disease detection via wearable sensors.