Adaptive probability scheme for behaviour monitoring of the elderly using a specialised ambient device

Winkley, Jonathan; Jiang, Ping

Abstract

A Hidden Markov Model (HMM) modified to work in combination with a Fuzzy System is utilised to determine the current behavioural state of the user from information obtained with specialised hardware. Due to the high dimensionality and not-linearly-separable nature of the Fuzzy System and the sensor data obtained with the hardware which informs the state decision, a new method is devised to update the HMM and replace the initial Fuzzy System such that subsequent state decisions are based on the most recent information. The resultant system first reduces the dimensionality of the original information by using a manifold representation in the high dimension which is unfolded in the lower dimension. The data is then linearly separable in the lower dimension where a simple linear classifier, such as the perceptron used here, is applied to determine the probability of the observations belonging to a state. Experiments using the new system verify its applicability in a real scenario.

Citation

Winkley, J., & Jiang, P. (2014). Adaptive probability scheme for behaviour monitoring of the elderly using a specialised ambient device. International Journal of Machine Learning and Cybernetics, 5(2), 293-307. https://doi.org/10.1007/s13042-012-0134-4

• Journal Article Type: Article
• Acceptance Date: Sep 14, 2012
• Online Publication Date: Oct 5, 2012
• Publication Date: 2014-04
• Deposit Date: May 19, 2015
• Publicly Available Date: May 19, 2015
• Journal: International journal of machine learning and cybernetics
• Print ISSN: 1868-8071
• Electronic ISSN: 1868-808X
• Publisher: Springer Verlag
• Peer Reviewed: Peer Reviewed
• Volume: 5
• Issue: 2
• Pages: 293-307
• DOI: https://doi.org/10.1007/s13042-012-0134-4
• Keywords: Hidden Markov model, Fuzzy system, Dimension reduction, Linear separation, Elderly monitoring
• Public URL: https://hull-repository.worktribe.com/output/373884
• Publisher URL: http://link.springer.com/article/10.1007%2Fs13042-012-0134-4
• Additional Information: The final publication is available at Springer via http://dx.doi.org/10.1007/s13042-012-0134-4