Location Prediction Using GPS Trackers: Can Machine Learning Help Locate Missing People with Dementia?

X25426605

Jan 22 2019  

JanuszWojtusiak ReyhanehMogharabnia

IoT Journal has just published Associate Professor and Director of Health Informatics at Mason’s College of Health and Human Services Janusz Wojtusiak, PhD, initial findings from his research which studied the possibility of using machine learning methods applied to data from GPS trackers to create individualized models that describe patterns of movement. These patterns can be used to predict typical locations of individuals with dementia, and to detect movements that do not follow these patterns and may correspond to wandering. The GPS trackers and data used in the study were provided by GTX Corp’s GPS SmartSoles.

Screen Shot 2019-01-28 at 9.43.23 AMSignificant number of people with dementia are at risk of wandering and getting lost. These individuals may get hurt, cause distress to families and caregivers, and require costly search parties. This study explores the possibility of using machine learning methods applied to data from GPS trackers to create individualized models that describe patterns of movement. These patterns can be used to predict typical locations of individuals with dementia, and to detect movements that do not follow these patterns and may correspond to wandering. Data from a sample of 337 GPS trackers were used. After pre-processing the data are used for iterative clustering, followed by classification learning. The number of clusters ranged between one (devices that always stayed “home”) and nine for devices with maximum mobility. The average number of clusters was 2.62. Models for predicting location achieved varying accuracy, depending on regularity of wearer’s schedule. The achieved average Area under ROC (AUC) is 0.778, with accuracy 0.631, precision 0.662, and recall 0.604. Unusual locations that potentially correspond to wandering incidents were identified by applying a secondary classification learning after filtering out data corresponding to normal movement.

 

GET FULL ARTICLE ACCESS HERE.