Time Series Data Processing for Classifying Wandering Patterns in People With Dementia

As the population ages, dementia has become one of the main health issues worldwide affecting the elderly. It is a disease related to the damage of the brain cells, causing memory loss, impairing of written and spoken communication skills, difficulty in performing previously routine tasks, as well as personality and mood changes. There is no cure for dementia, but if diagnosed correctly, providing the proper treatment and support allows for a better quality of life for those affected by this disease. People with dementia tend to wander, and a relationship between the wandering pattern and the level of dementia has been established. In this paper, two-time series techniques, the autocorrelation function and the partial autocorrelation function used in conjunction with the machine learning algorithms, including linear discriminant analysis, multivariate Gaussian, adaptive boost, and k-nearest neighbors, were evaluated to classify wandering patterns in people affected by dementia. The main contribution of this work is the use of time-series data techniques and machine learning algorithms to classify wandering patterns. The use of smoothing filters and time series feature extraction techniques, used in combination with ML algorithms, showed a very good performance, with an accuracy greater than 90%.