A predictive location-aware algorithm for dementia care

Some mentally impaired but otherwise physically healthy individuals may have a tendency of wandering or difficulties using public transportation. Intelligent assistive technologies which are able to learn the individual´s travel behavior and prompt anomalous events such as when the person deviates from expected destinations would greatly enhance the independence and safety of the individual and also reduce the stress on family members and caregivers. This paper presents the design and preliminary evaluation of a prediction model that is a critical component of a mobile and personal wellness management system for patients with dementia. We define a generic architecture of assurance systems aiming to ensure the safety and well-being of such patients and also develop an enhanced class of state predictor for location-aware applications catering to dementia care. With our new scheme, the prediction accuracy has reached up to 90% compared to the average 76% to 81% accuracy achieved by Markov, Bayesian network, Multilayer Perceptron, Elman net or the original state predictor with confidence estimator. Considerations for the development of future dementia care systems are also addressed.